請同意我們的隱私權規範，才能啟用聽新聞的功能。

（中央社東京22日綜合外電報導）支撐網際網路運作的程式語言PHP爆嚴重漏洞，全球企業與政府今天加緊腳步搶修，多國銀行、航空、鐵路與政府網站緊急暫停服務，駭客已嘗試攻擊達數百萬次。

路透社報導，PHP與HTML、Python同為關鍵的電腦程式語言。PHP開發團隊昨天針對潛藏多年、直到最近才被人工智慧（AI）揪出的漏洞發布修補程式，並呼籲使用者「立即更新」。

歐洲及其他地區政府昨天接獲消息後，已預防性關閉部分網站，網站營運商則忙著強化執行PHP的系統安全，不僅凸顯各國政府和企業有多依賴這種少有人懂的程式碼，也是全球仰賴PHP數十年以來，面臨最嚴峻的考驗之一。

微軟（Microsoft）執行長納德拉（Satya Nadella）今天在德國的活動上說：「我已經很久、很久沒看過這種狀況了。」他表示，這起事件顯示，隨著系統日益複雜，打造能夠安全且持續維護的軟體更加重要。

比利時資安公司Aikido Security共同創辦人勞倫斯（Madeline Lawrence）表示，漏洞細節在修補程式備妥前便已外流，此後陸續傳出攻擊，企業正和能利用AI發動攻擊的駭客賽跑。

雲端服務商Cloudflare執行長普林斯（Matthew Prince）昨晚在社群平台X指出，漏洞公布前10小時就已偵測到數千次攻擊嘗試，在漏洞公布後的45分鐘內，更暴增至數萬次，短短2小時內攀升至數十萬次，目前已達數百萬次。

綜觀全球，日本受到的衝擊尤其明顯，野村證券（Nomura）、三菱日聯金融集團（MUFG）、全日空（ANA）、JR東日本公司以及日本財務省、氣象廳、國稅廳都暫停部分網路服務；數據顯示，日本今年前9個月記錄到的網路安全事件，已超過去年全年。

法國電信巨頭Orange也預防性暫停旗下多家子公司部分以PHP建置的網站和應用程式；中國大型雲端服務商騰訊雲則呼籲客戶更新PHP版本，但沒有要求大規模關閉網站。

芬蘭國營資訊服務機構Valtori在前一天關閉政府公共網站後，已連夜安裝修補程式，並逐步恢復服務。

亞洲多國政府今天也下令針對PHP漏洞進行資安檢查。中國工業和信息化部呼籲各單位與使用者檢查系統，並升級至最新的安全版本；韓國則調高民間部門的網路安全警戒等級，並啟動緊急應變措施。

德國資訊商業通訊新媒體公會（Bitkom）安全部門主管庫倫坎普（Felix Kuhlenkamp）指出，許多較大型業者已完成更新，但複雜的資訊系統和小型機構可能需耗費更多時間。（編譯：張茗喧）1151023

黃仁勳批AI毀滅人類論：搏版面的不實廉價說法

請同意我們的隱私權規範，才能啟用聽新聞的功能。

（中央社柏林22日綜合外電報導）美國人工智慧（AI）晶片業巨擘輝達（Nvidia）執行長黃仁勳今天表示，有關AI可能毀滅人類的相關警告，是搏取新聞版面的「廉價」嘗試。

法新社報導，黃仁勳在他的歐洲行終點站德國首都柏林出席輝達主辦的1場會議時表示：「當你說人類文明將終結，首先這不是真的。」他直言，這種說法會製造新聞話題，但其實是「捏造」且「廉價」。

近來因發生AI曾在測試時躲過控制，引發其是否會威脅人類生存的激辯。美國AI開發商Anthropic的1名研究人員9月辭職，並強烈警告一些領先的AI實驗室正在「拿我們的生命做賭注」；其後Anthropic和同業OpenAI的負責人也皆呼籲放慢開發腳步。

透過 Google News 追蹤中央社

然而，黃仁勳提倡全速發展AI。他今天在柏林呼籲全球以「負責任的樂觀態度」來看待AI。

他說：「如果你關心未來、（關心）我們個人生命結束後的時代，如果我們在乎永續性的前景，那麼現在正是最佳時機：打造AI，不要試圖扼殺AI。」（編譯：張正芊）1151022

Colossal Biosciences’ Ben Lamm and Andruil’s Palmer Luckey had an idea to genetically engineer war dogs for the U.S. military, law enforcement, and other emergency response teams. That idea, on Thursday, announced its emergence from stealth as Warhound, armed with $200 million in funding from investors including In-Q-Tel, Sequoia, and 8VC. The round values the company at $1.2 billion.

These are dogs that perform tasks that are dangerous for humans or difficult for machines, like bomb detection, searches, and other specialized operations. In its launch announcement, the company said that more than 80% of the working dogs used by the U.S. military come from Europe and less than half of them are fit for service.

So not only is there a dwindling supply of working dogs for these tasks, but the U.S. has also become increasingly reliant on foreign nations for dogs that play an important role in the U.S. military and in law enforcement. That is key for what Warhound is trying to do. The hope is that Colossal can use its technology to engineer dogs with “characteristics suited to demanding missions,” the company said in a press release.

Lamm told TechCrunch he met Luckey about a year ago at a conference, where the Anduril co-founder brought up doing more for working dogs in the U.S. military. The two kept in touch, and Lamm did more research, learning about the breeding issues. Many dogs do not make it through training because of genetic dispositions for issues like poor eyesight or the high possibilities of cancer, hip dysplasia, or muscle atrophy.

Lamm believed dogs could be enhanced for health, longevity, and “for things like speed and agility or enhance olfactory sensors for bomb sniffing and whatnot,” he said. His team looked at the top-performing working dog breeds — the Belgian Malinois and German Shepherd — taking the cell lines from them and putting them in a biobank. From there, they will edit the genes, adding the ones that will make the dogs perform better while taking out the ones that won’t, to create “a new type of dog.”

Andruil, meanwhile, will provide software and hardware enhancement for items such as dog protection vests and eye goggles to assist the dogs and their handlers with matters like situational awareness and coordination.

He said to imagine, for example, bulletproof vests that have data sensors that give information back to the handlers so they can better understand what is happening.

“The idea is, if we make the dog safer, make it healthier, make it perform better, it’s going to perform better in the line of duty,” Lamm said. He added that Anduril’s additions aim to make the dogs safer and healthier outside of their genetic components, “on top of the dog, not in the dog.” He believes such enhancements will also protect the dogs’ human handlers better.

Luckey doesn’t own equity in the company and is only a partner, Lamm told TechCrunch, clarifying that Colossal owns equity in the business, which it supports with licensing its technology.

Colossal is currently known for its de-extinction efforts of animals like the dodo and the mammoth. As Lamm previously told TechCrunch, the company has been using AI, buying and building technology to help with gene cloning and the identification and reproduction of animal genes.

The first batch of dogs is about to be cloned and incubated and is expected to be born early next year, ideally entering service within the next two years, Lamm said. Another company owned by Colossal, Viagen, is partnering with Warhound to help clone the dogs. Lamm said Viagen’s tech has a high success rate with cloning and Colossal currently uses it for its program to help protect endangered species.

The startup doesn’t know yet how much it will charge for the dogs but expects to sell to parties like first responders, the Department of Defense, military service groups, and police forces. When asked about government approval, Lamm said the company has briefed many government groups about the business.

In terms of the responsibility and safety of civilians when it comes to releasing genetically enhanced dogs, Lamm said the dog will be trained like any military working dog and will be managed by whoever buys it.

The company said it is working under the guidance of the American Humane Society to help certify Warhound’s training program for the newly created dogs that will enter service. The hope is also that when these new military dogs retire, they will enter the Society’s Military Working Dog Reunification Program, which provides a home for retired military dogs — or another program the society offers to help dogs.

Spotify had a strong third quarter, reaching 788 million monthly active users and generating nearly €5 billion (around $5.6 billion USD) in revenue. But the more interesting takeaway from its earnings call was how the company sees people listening to music in the future, and why the rise of AI agents could be good news for its business.

Co-CEO Gustav Söderström thinks the music streaming giant is particularly well-suited to this shift.

“Spotify is in one of the few categories that actually stands to thrive in the agentic age,” Söderström tells investors. “This is because agents excel at replacing and taking care of things people would rather not do. People love getting rid of chores. But entertainment is the opposite. Nobody asks an agent to ‘summarize a song’ for them.”

Instead of opening Spotify, searching for an artist, and building a playlist themselves, listeners could ask an AI assistant to do the work for them. They might want music for a run, a podcast about something they’ve been discussing with an AI, or an audiobook recap that picks up exactly where they left off.

Spotify wants to be the service those agents turn to, letting people discover music and interact with Spotify through conversations rather than just its own app. Notably, Spotify was the first music service to integrate with Meta’s Muse assistant, and it also integrates with ChatGPT and Claude.

“Our goal is to be the media service that you use from any agent, really turning the growth of agents into a tailwind rather than a headwind. That is exactly what we are seeing now. We are seeing really good adoption of Spotify inside these agents. While other companies are maybe trying to figure out what to do and many are leaning out, we are actually leaning in and participating in the growth instead of standing by the roadside and looking at it,” said Söderström.

The market is already showing signs of friction over how agents access third-party services. Amazon, for example, has blocked Meta’s Muse from shopping on its platform, illustrating how major players may restrict outside agents when questions arise around access, security, privacy, or commercial control.

Spotify’s pitch isn’t just that AI should just make entertainment easier to find. It should make the experience more personal, too.

Söderström describes Spotify today as “an intelligent media service encapsulated as an agent, whether you use it directly through our application or via another agent…Spotify is now an agent that you can not only click on in our app, but talk to, ask questions of, and instruct. We can do things today that no other media services can do,” he said.

“For example, you can ask Spotify to go through your listening history song by song back to 2008, picking your true favorites, beat-stretching them by speeding them up or down, and then sorting them by key and BPM. Then you can ask it to mix them together into the perfect running playlist at your exact cadence,” Söderström explained.

Spotify has years of listening history to draw on, including the artists people return to, the songs they skip, and the music they discover. That information can help it make recommendations that feel more personal than a generic playlist generated by an AI assistant that doesn’t know much about someone’s tastes.

However, Söderström also emphasized that Spotify isn’t sending its listener data to partners. Instead, the agent asks Spotify for something, and Spotify uses its own systems to reason about the request and choose what to play.

Fusion startup Helion has been awarded $175 million in federal funding to research, develop, and manufacture advanced semiconductors and power electronics for its fusion power plants, the company announced on Thursday.

Currently, those devices lose about 5% to 10% of the electricity that flows through them. While those losses seem small, they can compound themselves in a complex system like Helion’s.

“If we can cut that loss in half, that’s actually twice the power output of the fusion system,” David Kirtley, co-founder and CEO at Helion, told TechCrunch.

Helion has been racing to complete its first power plant, Orion, in time to meet a 2028 deadline to provide Microsoft with electricity from the grid. Funding from the award will help Helion build automated semiconductor module, packaging, and testing capacity to supply the fusion machine that’s currently under construction.

The new production line, to be built in Washington State, will be the first in the U.S. to make a class of semiconductor that is used not just in fusion power, but in manufacturing, wastewater treatment, medicine, and defense.

“There are no U.S. manufacturers right now that do it,” Kirtley said. “This is absolutely enabling going from zero to 100 for advanced, high-voltage, and pulsed semiconductors here in the United States.”

Fresh funding

Helion Energy’s facility in Everett, Washington — Helion Energy plans to open the world’s first fusion power plant by 2028 and supply Microsoft Corp. with energy from it soon after Image Credits: / Getty Images Mike Kane/Bloomberg (opens in a new window) / Getty Images

The three-year award was made by the Department of Commerce under the CHIPS Act, a Biden-era law that seeks to bolster semiconductor module research, development, and manufacturing in the U.S. Alongside the federal funding, Helion also announced a fresh $50 million private investment, though it didn’t disclose the investor. In total, Helion has raised more than $3.2 billion, making it the second-best-funded fusion startup.

More than $100 million will be set aside for semiconductor and advanced packaging R&D in Washington State, while approximately $80 million will support R&D efforts in Vermont for advanced semiconductor tooling, design, and process development with GlobalFoundries, Kirtley told TechCrunch. Separately, GlobalFoundries has received more than $80 million from the federal government to advance the manufacturing of gallium-nitride semiconductors, a type that is widely used in power electronics.

Kirtley acknowledged that the R&D funded by the award won’t make its way into Orion in time for the launch — federal contracts can take months to negotiate — but he said they promise to help improve the efficiency of subsequent Helion power plants, which would improve their competitiveness in cutthroat electricity markets.

Efficiency is everything

Semiconductor efficiency is key to Helion’s approach to fusion, which harvests electricity directly from the fusion reactions rather than using them to heat steam to spin a turbine.

The startup is developing a reactor known as a field-reversed configuration. Fusion fuel gets injected into both ends of an hourglass-like chamber, where it’s heated to form a plasma. The plasmas are then accelerated toward each other using magnetic fields. When they collide, another magnetic field compresses the ball further, raising the temperature to over 100 million degrees C.

Deuterium and Helium-3 are heated, then accelerated through magnets, compressed, and captured as inductive current Image Credits:Helion

At those temperatures and pressures, atoms in the fusion fuel start fusing, releasing energy in the process, and pushing the plasma outward. This force pushes against the magnetic field, and that energy is captured by the magnets and capacitors and turned into electricity — think of how a hybrid car or EV recharges the battery by using electric motors to slow the vehicle.

Helion says that approach, known as direct energy conversion, will help future reactors produce more power than they need to operate, a milestone that any fusion power plant must reach if it is to be commercially viable. Efficient power electronics are a key part of the process.

“One of the enabling parts of our technology that allows us to do this direct electricity recovery — putting electricity into fusion and at a very high efficiency, recovering it at high speed — is the modern semiconductor,” Kirtley said. “We’re going to take that to the next level here, increase the efficiency.”

The energy sector isn’t the only place where advanced power electronics could have an impact. Electric vehicles represent the broadest use today. Consumer electronics could benefit, too, though there the size and power demands are lower than Helion’s.

The defense industry is likely to be another beneficiary. Energy weapons have garnered fresh interest as drones have swarmed battlefields and overwhelmed existing air defense weapons. Bill Frauenhofer, who leads semiconductor investment and innovation at the Department of Commerce, acknowledged in a press release that the award will help “convert and deliver power to critical applications in defense, grid transmission, and advanced energy generation.”

Despite the potential for defense contracts, Kirtley said Helion remains focused on putting power on the grid. “Domestic energy supply, national security through energy dominance, and energy security — that’s our main focus,” he said.

In Brief

In a sign that crypto and blockchain-focused investing has receded in importance, Andreessen Horowitz announced Thursday that it is renaming its digital asset unit from a16z Crypto to a16z Frontier.

The practice, helmed by Chris Dixon, has raised nearly $10 billion across five funds since launching in 2018, backing some of the sector’s biggest wins, including Coinbase. Yet for years, a16z and rival venture firms struggled to push crypto’s utility beyond financial rails, trading applications, and speculative assets.

Under the new name, the team is broadening its mandate to focus on emerging technologies, be it AI, robotics, or “new kinds of networks” connecting humans and machines. Yet, the group isn’t abandoning crypto. Dixon wrote in a blog post that blockchain technology could “give new networks native economic capabilities.”

Alongside Dixon, a16z Frontier’s general partners include Eddy Lazzarin and Guy Wuollet. Meanwhile, Ali Yahya, who joined the crypto strategy as its first investor in 2017, has left the firm, according to The Information.

In Brief

OpenAI is reportedly close to securing another mega round of private funding, rather than plowing ahead with an IPO. The AI lab is working on raising up to $30 billion at a $1.25 trillion pre-money valuation in a round led by Coatue and investors in UAE, sources told CNBC.

This follows a similar report by Bloomberg a month ago which said the AI lab was working on raising $30 billion at a $1.4 trillion valuation. Terms are still being negotiated, both outlets report, so they could ultimately change. CNBC reports that the UAE funding could also be coupled with a deal to secure more computing capacity for OpenAI from the UAE, too. OpenAI declined comment.

Should OpenAI command a valuation of more than $1 trillion with this round, it would break records as the most valued private company ever. That valuation would also raise the bar even higher for OpenAI’s forthcoming IPO. OpenAI filed confidential paperwork for an IPO in June, after its last raise in March valued the ChatGPT maker at $852 billion post-money.

That high valuation could also have ramifications for Anthropic’s IPO, which many believe is closer to materializing. (Rumors suggest we could see Anthropic’s public S-1 filing any day now.) The latest reports have Anthropic seeking a $1.5 to $2 trillion valuation for its public debut, CNBC reports. Anthropic raised $65 billion at a $965 billion valuation in May in what many believed at the time would be its last big round before an IPO. It also filed confidential paperwork in June.

EV maker Rivian has acquired Third Dimension, a U.K. startup founded by veterans of the gaming and autonomous vehicle industries that uses AI to turn images and videos into lifelike 3D worlds.

Rivian isn’t getting into gaming, but it does want to leverage Third Dimension’s technology to turn real-world data into 3D worlds that can accelerate the development of its advanced driver-assistance features and the self-driving system for the robotaxis it plans to supply to Uber.

Terms of the deal were not disclosed. The 15-person team at Third Dimension, most of whom are based in the U.K., will be absorbed into Rivian’s autonomy and AI team led by James Philbin. The acquisition also gives Rivian an opportunity to expand its AI hub in London, Philbin said in an exclusive interview.

Third Dimension, which was founded by CEO Tolga Kart and Piotr Sokólski in 2024, raised $7 million in seed funding from venture capital firms, including Abstract Ventures, MVP, Soma Capital, and Solari Capital. While the company’s technology was built to speed up video game development, the founders saw other potential applications, including defense and autonomous vehicles.

Now, under Rivian, they’ll be focused on automated driving.

“The team is excited to be able to focus on one problem,” Kart told TechCrunch in the interview. “I think as a small company, having to serve many, many different masters is very hard, and so we’re very excited to come work on a problem that has such an amount of scale, where our technology makes a true difference.”

Simulation is a critical tool for autonomous vehicle developers, who use it to train and validate self-driving systems. Historically, these simulators were built on rendering or video game engines and often relied on synthetic data. These virtual environments might look like the real world, but a gap between the two remains, Kart said. The workaround typically requires feeding the simulators ever-larger amounts of data in hopes of closing that quality gap.

Third Dimension’s technology uses AI to match the physical world, which when combined with Rivian’s reams of data could turn into a valuable simulator.

Rivian collects driving data from tens of thousands of R1S pickup trucks, R1S SUVs, delivery vans, and R2 SUVs on the road today — essentially any vehicle whose owner has enabled external camera sharing. Rivian uses that data, which is aggregated and anonymized, to build training models for driving. According to Philbin, Rivian gets more than 400,000 sequences uploaded from its customer fleets per day.

Third Dimension’s tech will allow Rivian to simulate complex driving conditions without having to wait for rare real-world edge cases to occur, Philbin said.

This story is developing and will be updated.

The Browser Company is sunsetting its design-forward Arc browser on December 22, 2026, and directing users to Dia, its AI-first browser. The company announced the decision on Thursday, citing the rising cost of keeping a browser secure in the age of AI.

“Keeping a browser secure takes constant, dedicated work that is only increasing in the age of AI,” the company wrote.

“Dia was built from the ground up to meet today’s security standards, so that’s where our engineering and security teams will focus moving forward.”

The wind-down is staged into three parts. As of Thursday, new Arc accounts can no longer be created. On November 23, Arc Search on iOS and Android will lose its AI features, including “Browse for Me.” Users will no longer be able to sign in or sync tabs, though the app will continue functioning as a basic local browser.

On December 22, Arc will stop working entirely on Mac and Windows. Accounts will be deactivated, and personal data will be deleted within 30 days. Arc Search will also be removed from the Google Play Store.

While The Browser Company claims Dia now surpasses Arc in speed, stability, and security benchmarks, user reactions on social media were largely disappointed, with many reviving requests for the company to make Arc open source.

But CEO Josh Miller previously pointed out that Arc is built on the Arc Development Kit (ADK) — a proprietary internal SDK that is also the core engine powering Dia. Open-sourcing Arc would mean giving away Dia’s intellectual property.

“While we’d love to open-source Arc someday, we can’t do that meaningfully without also open-sourcing ADK,” Miller previously noted. “And ADK is still core to our company’s value.”

For users making the switch, Dia will carry over equivalents of several core Arc features. Users can still swipe between different work and personal environments using Dia’s “Profiles” (formerly Arc’s Spaces). Arc’s signature vertical tabs will remain as an optional sidebar layout, while Folders have been rebranded as “Pinned Tab Groups.”

Passwords, bookmarks, extensions, and history will seamlessly transfer to Dia. Arc features will remain free in Dia, with AI optional.

The shutdown officially closes a chapter that began in 2025, when the startup stopped active feature development on Arc. The Browser Company — which was acquired by Atlassian for $610 million in 2025 — conceded at the time that while early adopters loved the browser’s novel interface, the “novelty tax” and complexity limited mainstream adoption. Arc was fiercely beloved by a dedicated group of enthusiasts, but it never quite became the mass-market Chrome alternative its makers once envisioned.

The AI era has changed how fast startups can grow. Startups can grow so fast that they hit high annual recurring revenue (ARR) numbers, only to crash and burn a few months later. At this year’s TechCrunch Disrupt, Lovable go-to-market lead Luke Arno, Ben Broca of the AI platform Polsia, and Theory Ventures’ Tomasz Tunguz gave founders some tips on how to set their companies up for success in those first 90 days.

Broca, whose company Polsia uses AI agents to build and run businesses with almost no human staff (Broca says he is its only employee), said that product-market fit is one of the hardest things to find right now, especially in an era when so many products can easily be copied by competitors. In response, Arno said that durable go-to-market strategies are those that can be reliably repeated.

“The most critical thing for founders or earlier-stage companies to do is develop learning density or pattern recognition for where your customers are,” Arno said. “Don’t rush to build out a top-down go-to-market motion if you’re not there yet.”

That’s why, in those first 90 days after launching a commercial product that’s ready to sell, founders should not mistake early, fast revenue for a sign that the startup is ready to scale, Arno said.

Once a startup finds use cases that work, it will only be ready to build out a sales team and go for growth after it can attract new customers by repeating what has worked. If every sale requires new tactics, the startup hasn’t truly hit product-market fit.

“If I bring on 10 new customers in 10 different ways to solve 10 different problems, sure, I brought revenue in, but I don’t have a way to go scale a go-to-market,” Arno said. “If I bring three customers in with a similar motion to solve a similar problem, now I have high confidence that I can go scale.”

Lovable, the AI app-building startup, set a lot of those early benchmarks with how fast it grew, and now founders everywhere are feeling the pressure to do the same. Lovable is valued at more than $13 billion by its investors, says its ARR has passed $600 million, and has raised more than $700 million in funding to date.

“The expectations for overall execution speed are absolutely tremendous,” Tunguz said.

But Tunguz said growth like Lovable’s is still the exception, though founders should still move fast. The pressure they feel is real, he said, because far more money flows through the startup ecosystem at every stage than when he began investing in 2008. Today, a seed deal brought in by one of his partners could just as easily be a $5 million raise, a $50 million raise, or a $500 million raise.

The winners all focus on one thing: understanding their customers well enough to know what will keep them around even after any hype subsides.

“Community building is very important and being close to customers is very important,” Broca said. He said founders can send emails and throw events, and he personally likes to talk to a lot of his customers: “The general rule is always to be talking to customers.”

Concerns have swirled over the access that consumers are increasingly entrusting to AI agents. Now, security services are trying to catch up.

Password manager 1Password on Thursday launched a new tool by which users can allow agents to log into their personal accounts on their behalf, without the agent ever having access to a user’s credentials.

The new platform, dubbed 1Password for Agents, accomplishes this by creating a system of communication between the agent, the user, and the password manager. When an agent attempts to log in on a user’s behalf, a unique digital key is sent to 1Password to confirm the agent’s digital identity. The user is then sent an alert about the login attempt and can personally confirm the request. 1Password then signs in the agent, filling in the credentials on the digital assistant’s behalf, which allows it to log in and complete whatever task is at hand.

1Password for Agents allows users to choose their agent of choice — whether that’s Claude Duet from Anthropic, OpenAI’s Dots or ChatGPT Work, Grok Bot, Meta’s Muse, or agents from Town, which is an AI assistant startup. The company says it is working with other companies to expand the service.

“Handing over a password, secret, or credential is a common shortcut to an agent, but can leave people with sensitive information in places they can no longer control as a copy gets saved for the next time,” the company said in a blog post. “1Password for Agents gives people a trusted way to delegate tasks to personal AI agents while keeping control of the passwords, secrets, and credentials those tasks require.”

Online agentic activity has increasingly unsettled major institutions. Indeed, 1Password’s new platform is launching on the same day that Axios reports that U.S. banks have begun to detect agentic activity inside user accounts.

This isn’t necessarily a case of rogue agents pilfering accounts but, rather, an example of how consumers are increasingly allowing automated services to act on their behalf, including on financial matters. However, the potential for agent-related problems is still large. The big issue appears to be that most banks cannot tell if the agentic activity is tied to user requests or not.

Bankers interviewed by the outlet described the situation as “horrifying,” and a Chase spokesperson interviewed by Axios additionally told the outlet that customers “should not give their usernames and passwords to anyone.”

Design platform Canva released its first image generation model, called Canva Craft, developed by its research team to compete with same-category models from Google, OpenAI, ByteDance, and Meta.

The company said its model works in two stages. In the first stage, it interprets users’ prompts and creates a plan with details like spatial relationships of elements, relative scale, visual hierarchy, and information for text generation. In the second stage, its model generates an image of up to 2K resolution in a possible 30 different aspect ratios for various platforms.

Canva said that this approach allows it to create images with accurate details of objects and legible text. Plus, users would be able to make precise edits without changing other parts of the image.

The company noted that rather than using photographs, it used designs to train this model using human-designed templates from Canva’s library, including posters, social posts, presentations, infographics, menus, and typography-rich layouts.

That means images generated with Canva Craft could be converted into layered designs using Canva Magic, the company’s image-to-design transformer. Last year, the platform released its own model that understands designs and layers.

Canva Craft is available on the company’s platform to all users, who can access the model from the Dream Lab tab. The model will also be available through the Canva API. Plus, the company plans to support image generation in third-party chatbots like ChatGPT and Gemini through Canva’s MCP.

Canva has used a bunch of image models for generation in its platform, so it’s likely that the new models will help save API costs for the company. It said that besides costs, the new model also allows the platform to have better quality and speed control.

In 2024, Canva acquired media generation platform Leonardo.ai. It said that the researchers from Leonardo contributed to the new image model.

Canva said that it is developing two cost-efficient versions of the model, called Craftfast and Craft-ultrafast, which will be released soon.

The company was on track for $4 billion in revenue last year. However, it cut its growth forecast for 2026, causing its valuation to reportedly plummet from $42 billion to $35 billion, according to external analysts.

In Brief

Former OpenAI science lead Kevin Weil has launched Warp Sciences, a new AI startup focused on automating lab work for drug discovery. The company, which trains robots to make and test promising compounds identified by researchers, has raised $215 million at an $865 million valuation, according to Weil, who announced the startup and funding on X.

The round was led by Lightspeed, with participation from IVP, Premji Invest, and Lux Capital.

For years, biotech companies have outsourced the most labor-intensive work of synthesizing thousands of variations of molecules to contractors in China, such as WuXi AppTec. With Warp Sciences, Weil hopes to reshore that work and insulate the entire drug development process from geopolitical supply-chain risks.

Wearing cameras and sensor gloves, the startup’s human chemists will train robots to perform various lab tasks. By automating wet-lab work, Warp Sciences aims to accelerate drug discovery using robots that can operate around the clock.

Microsoft’s AI diagnostic system correctly diagnosed complex medical cases in a simulation more than twice as often as generalist physicians, according to a study published Thursday in Science. Mustafa Suleyman, Microsoft AI’s CEO, called the result a step toward “medical superintelligence.”

The system, dubbed MAI-DxO (DxO for short), coordinates several AI agents that act like a panel of doctors. They question a patient, order tests, and debate a diagnosis.

In a simulation using complex cases from the New England Journal of Medicine, it was correct 83% of the time. Primary care doctors and general internists, working individually with reference materials, were correct 36% of the time. One note: Cases this complex are usually handled by specialists or physician teams, who weren’t part of the comparison.

“There is something very substantial and exciting here,” Dominic King, vice president of Health at Microsoft AI, told TechCrunch. “The biggest challenge in healthcare delivery at the moment is the growing mismatch between demand and supply of medical expertise.”

In publishing these results, Microsoft joins Google and OpenAI in claiming that its AI can match or outperform doctors.

The study’s own data suggests that much of the improvement comes from more powerful underlying AI models, with Microsoft’s orchestration system adding gains on top.

Microsoft tested MAI-DxO with models from several companies, including OpenAI, Google, Anthropic, Meta, and DeepSeek, but its best results came with OpenAI’s GPT-5 and GPT-5.5.

Accuracy improvements delivered by DxO (no budget constraints) on CPC over models

prompted with baseline prompt across different large language models. Image Credits:Microsoft AI

Microsoft measured DxO on cost as well as accuracy, meaning the estimated cost of running the AI on each case. On its own with a short, basic prompt, GPT-5 scored 78.9% on complex cases at about $8,400 per case. DxO cut that cost by about $3,000 but raised accuracy by only about four points. On emergency department cases, there was little improvement in either accuracy or cost improvement between GPT-5 on its own and with the assist from DxO.

With GPT-5.5, a newer model released in April 2026, the results split. On complex cases, DxO increased GPT-5.5’s accuracy score from 72% to nearly 87% at a slightly cheaper cost. However, the gains once again didn’t show up on emergency department cases. Both (with and without DxO) scored about 85% at a similar cost.

DxO helps most on complex cases and with less capable AI models. Gemini 2.5 Flash, for example, scored 46.4% on its own but nearly 69% with DxO. Even so, the gap between the weaker and stronger models was larger than the boost DxO provided.

Microsoft’s researchers said that accuracy gains naturally shrink as models approach the theoretical maximum. “The DxO orchestration helps with the broader challenge of optimization both accuracy and cost,” a spokesperson added.

Suleyman, who leads Microsoft’s effort to build its own frontier AI, told TechCrunch that the company is working with the Mayo Clinic to train a foundation model for healthcare, with release planned for next year. He believes this model will enable the company to “make really enormous leaps here,” and says DxO is part of the foundational infrastructure Microsoft is building to see real breakthroughs in healthcare.

“I really believe that in 2027, we will be on the verge of a coding style breakthrough for medical records,” Suleyman said. “I think we’re going to be able to stream highly accurate predictions about how a medical record might unfold, and obviously that kind of foresight for a clinician is gold dust.”

Suleyman has spent his career trying to use AI to solve medical issues. He was previously a co-founder of DeepMind, which was acquired by Google in 2014. The following year, Google DeepMind acquired King’s startup Hark, a clinical task management app, and the two of them launched DeepMind Health together. Suleyman joined Microsoft to lead its AI unit in 2024, and King followed a few months later.

“We’re just true believers,” Suleyman said.

He said while the use of generative AI in healthcare is still early, he expects healthcare models to get “an order of magnitude better in capability within the next 24 to 36 months.”

“As we’ve seen multiple times in the last five or six years, the progress happens very quickly once you have a flywheel,” Suleyman said. “A flywheel means we have a model in production which we are iterating with real-world usage in order to identify gaps in the model and then patch them and improve them. That’s what’s happened in coding. It’s what happened in chat before that. And I think next year it’s going to start to happen for medical records and for clinical decision support.”

Through a new partnership with Beth Israel Deaconess Medical Center in Boston, Microsoft will evaluate the clinical reasoning capabilities of DxO across different hospital settings.

“This technology is ready for real-world evaluations, and you can imagine the complexity and diversity of issues that come from a Boston emergency department on a Friday evening,” King said.

He explained how the study will work: Patients will get standard treatment, including triage, time with a specialist, treatments, investigations, and decision on whether it’s safe to discharge. In parallel, those who consent will also engage with DxO. Microsoft and its partners will compare DxO’s recommendations against the human doctors’.

Google published a clinical study alongside Beth Israel in The Lancet this month to evaluate Google’s AMIE conversational diagnostic AI in real-world urgent care visits, while human physicians monitored the interactions in real time.

“We think this will be the biggest study of its kind,” King said. “It will be across not just emergency settings but urgent care environments, and will give us very compelling evidence by the end of the study period around the ability for these systems to be deployed at greater scale in the future.”

Microsoft’s study in Science coincides with a manifesto of sorts around what it will take to build medical superintelligence, defined as “the capacity to surpass the strongest contemporary human-led health systems in medical reasoning and coordination across the full arc of care.” Microsoft hopes to move beyond benchmarking individual tasks — like answering medical exam questions and generating diagnoses — into integrating with clinicians and healthcare systems for better outcomes.

“This is the kind of technology that everybody in the world wants, and that’s why we’re committed to doing everything we can to accelerate its introduction to the world,” Suleyman said, adding that this focus could be a differentiator for Microsoft compared to the other frontier labs.

Microsoft already sells dictation and radiology tools to clinicians and sees DxO as an expansion to those efforts.

“We have to have the absolute number one state-of-the-art medical clinical decision support products over the next few years,” Suleyman said.

In Brief

Elon Musk has invested approximately $1 billion in the newly formed media company Skydance, according to a new report from Semafor. News of the investment comes a few weeks after Paramount completed its $110 billion acquisition of Warner Bros. Discovery to create the combined company.

Co-CEOs David Ellison and Ynon Kreiz are tasked with bringing together two major streaming platforms, Paramount+ and HBO Max, along with a portfolio of networks, including CBS, CNN, MTV, TBS, Comedy Central, and Food Network, under a single corporate umbrella.

Semafor reports that Musk’s investment was made around the same time as equity contributions from other Skydance investors, including board member Laurene Powell Jobs. The investment will provide a much needed financial boost as the company seeks to cut costs and lift its stock price, the report notes.

Skydance has already committed to reducing costs by $6 billion over the next three years.

Musk has publicly praised the Ellisons in the past and has close ties with Larry Ellison, whom he appointed to Tesla’s board of directors in 2018. The Oracle founder invested $1 billion to help Musk purchase Twitter in 2022 and also invested in Tesla in 2018.

Skydance did not immediately respond to TechCrunch’s request for comment.

In Brief

The bad week for cybersecurity just got worse for thousands more companies, who were spied on by Chinese hackers following the breach of IT management and remote access giant ConnectWise.

Security researchers at Microsoft say the China state-aligned hackers broke into ConnectWise’s cloud backup service and stole application credentials that are used to establish data links between ConnectWise and its corporate customers, who rely on the cloud service to back up their company’s data.

By stealing the credentials, the hackers could log in to the customer’s Microsoft-hosted environment as if they were an authorized user and access their emails and other files.

According to Microsoft, the hackers used AI agents to rapidly hack into around 2,250 organizations at a rate of almost 3,000 attempted sign-ins per minute. Microsoft said this speed gives defenders “little time to detect and respond.”

The speed of the mass-hacks was attributed to the use of AI agents, specifically ARTEX, an open source AI framework that works with AI models to stress-test cyber defenses. The same AI software was used in a recent compromise of South Korea’s financial systems, many of which had weak security and lacked multi-factor authentication.

ConnectWise said it’s paused access to its cloud backup system and is investigating the breach. As of Thursday, the company said it was working on restoring the service.

This year, TechCrunch Disrupt showcased the incredible talent and groundbreaking ideas of our 2026 Startup Battlefield 200 cohort.

Out of thousands of applications, we selected 200 of the most promising startups, each bringing unique innovations to their respective industries. The competition culminated in an electrifying event where these startups had the opportunity to pitch and demonstrate their solutions live over three days.

From the top 20 finalists, TechCrunch editorial selected the top five companies that then battled it out for the $100,000 equity-free prize and the coveted Disrupt Cup. The well-deserved win went to Intero Biosystems, with a strong runner-up, Apate.

Among the Startup Battlefield 200 were many industry-defining companies exhibiting and pitching on the Showcase Stage. Here are the standouts:

Best Showcase Stage pitch by industry group

Built World

Lola Vision Systems provides a full-stack AI platform delivering sensor fusion, secure communications, and real-time inference on today’s hardware, with purpose-built silicon on the horizon.

Sustainability

Krokos Bio is growing nature’s most valuable ingredients, in a way that is cheaper and more reliable, starting with saffron.

Enterprise Group 1

Rein Security is the security layer for custom-built enterprise agents — the production AI agents running businesses worldwide.

Enterprise Group 2

Lumai is the optical compute company addressing scalable AI in the inference era.

Health

Uhura Bionics has developed a bionic larynx for singing, laughing, and translating emotions.

Consumer

Maka is delivering developmentally curated shows parents trust and kids love — powered by AI.

Best Booth

Legali AI is providing a trauma-informed AI legal companion and ecosystem for the middle class and AI-native firms, with a focus on women who’ve faced abuse. Their all-pink, female-focused booth was a standout on the Expo Floor. Visitors would write empowering messages to women on sticky notes and place them on Legali AI’s banner while learning more about the wonderful work they’re doing. Founder Ira Puteri also made a point of coordinating different pink outfits each day. Love the commitment!

Spirit of Disrupt award

The Spirit of Disrupt award unanimously goes to Shogher Atanesyan, founder and CEO of Ardook Data. Her journey to the Startup Battlefield 200 is inspiring, with her first exposure to Disrupt as a volunteer back in 2023. She had reached out to the Battlefield team last year, with a simple request to make “Social Impact” a stand-alone category in the Event App. We were happy to oblige, and just a few short months later she applied for the competition. She made it in on her own merits, as the review team loved the concept of capturing real-world household training data from a willing and paid caregiver network. The data is then sold to physical AI and robotics companies in need of it.

Atanesyan may be the most selfless founder we’ve seen in five years of managing the Startup Battlefield 200. Specifically, what she did for visitors to San Francisco. She organized dozens of student volunteers to help these startups manage their booths, coordinated numerous items shipped to her house to circumvent steep shipping costs, and hosted a party for these founders at her house. Additionally, she consistently helped founders who had questions in our community Discord and weekly calls.



The Startup Battlefield is more than a pitch competition; it’s an opportunity to build relationships with investors, potential customers, and fellow founders to support each other on their entrepreneurial journeys.

As we celebrate the achievements of this year’s winners, we are excited to see how they will shape their industries and drive innovation in the months and years to come. Congratulations to all the participants of the Startup Battlefield 200, and a special thank-you to our judges and sponsors for their support in making TechCrunch Disrupt 2026 a resounding success.

Uber and OpenAI announced an expansion of their partnership on Thursday, with the ride-share giant telling TechCrunch that its entire global workforce — some 30,000 people — will now be using ChatGPT Work.

ChatGPT Work was introduced in July as an enterprise-focused AI agent that corporate teams could deploy for various workflows.

Uber originally piloted ChatGPT Work with 1,000 employees earlier this year but has now scaled up its usage to the entire company.

That earlier pilot was deployed to workers from across Uber’s various departments, including finance, operations, and product. Indeed, a majority of the early adopters — some 80% — were non-engineers, Sumanth Sukumar, Uber’s VP of engineering, told TechCrunch.

“They used it for a wide range of tasks: analyzing complex data from our operations to identify trends and develop recommendations; localizing training materials for different markets and languages; and building personalized interactive dashboards to manage their daily work,” he said.

Yet while the ride-share giant is now going all in on ChatGPT, it has not settled on exactly how employees will use it. “Our approach is to make powerful AI tools available across Uber while giving teams room to experiment,” Sukumar said. “For example, earlier this year we paired 30 AI engineers with teams across Uber and gave them two weeks to automate key parts of their job. Those ‘AI pods’ showed us how much opportunity there is when you combine technical expertise with a deep understanding of a particular job.”

Uber didn’t disclose how much it’s spending on the new rollout of the chatbot. Instead, Sukumar said the company will track how employees use the tool. “We’ll look at how employees are using the technology, the kinds of workflows they develop and how it helps them work more effectively. We’ll evaluate those outcomes and use what we learn to guide how we deploy AI across the business,” he said. Uber says OpenAI worked closely with the company on integrating the chatbot.

OpenAI has moved quickly to scale up its enterprise business as it competes with its chief competitor, Anthropic. Once known largely as a consumer brand, the ChatGPT creator has swiftly gained a loyal following with business customers, while other AI companies have struggled to keep up.

Tesla says it is preparing to launch its shiny gold Cybercab in new markets outside its home of Austin, Texas, in a bid to “literally take over every part of terrestrial transportation.”

“We’re starting with getting people from one place to the other, but we want to basically own transportation of any kind on Earth,” Chief Financial Officer Vaibhav Taneja said on the company’s third-quarter earnings call Wednesday.

Tesla has previously shown a concept version of a multi-passenger vehicle it calls the “Robovan,” but it’s unclear if Taneja’s bombastic comments mean the company is working on other types of transportation.

What was clearer on the call is that Tesla’s Cybercab expansion will begin in Dallas next month and spread to other cities by the end of this year. Tesla already operates robotaxi versions of its Model Y SUV in small fleets in Dallas, Houston, and the Florida cities of Miami, Orlando, and Tampa. The company also said on its earnings call that it will start selling Cybercabs to fleet operators by the end of this year.

Broader expansion of the Cybercab could set up a real fight with Waymo, which is currently the market leader for robotaxis in the United States, with more than 4,000 of them deployed across more than a dozen cities. Tesla has argued that its Cybercab is cheaper and easier to build, which should allow it to undercut Waymo and other robotaxi operators on price. But the company has to reach real scale outside Austin in order to prove that point.

Tesla said it’s ready thanks to a dramatic increase in the number of miles its self-driving cars have traveled in recent weeks. While that may be true, Tesla was mum on Wednesday about how much money the overall robotaxi effort is bringing in. Tesla CEO Elon Musk said the new ride-hailing business won’t be noticeable in the company’s overall financial figures until the second quarter of 2027, “and probably a big deal by Q3 or Q4 of next year.”

Tesla also made a change to its letter to shareholders that further obscures what little visibility there was into its robotaxi business.

In Tesla’s second-quarter letter, the company included a chart for “Cumulative Paid Robotaxi Miles.” A close read of that chart showed a 36% decline from the first quarter of the year. In the new letter released on Wednesday, Tesla changed that chart to show “Cumulative Unsupervised Robotaxi Miles” — an indicator of how far the fleet cars are driving, whether or not they have paying riders on board.

There wasn’t much more clarity on the call, either. When asked about paid robotaxi miles traveled, Taneja said the company is “seeing utilization improving,” but didn’t offer specifics.

Instead, Taneja kept most of his comments generally focused on Tesla’s plans to scale the Cybercab. That included praise for Tesla’s engineering team for creating a vehicle that, according to the manufacturing team, is very easy to build — and build quickly.

“They think this is one of the easiest products for them to scale in terms of manufacturing,” he said. “So scaling would not be that much of a challenge.”

What might be a challenge, Taneja said, is that Tesla needs to make sure it has “all the regulatory approvals” so it can “start deploying the Cybercabs as fast as we can produce them.” That echoes many previous comments from Musk, who has long claimed that “regulatory approval” was the biggest barrier to Tesla’s robotaxi plans, not technical progress.

Musk has always been a bit hazy about what approvals, specifically, he felt were needed. The company has run into a regulatory hurdle this year, though. In September, just hours after Tesla formally launched its Cybercab in Austin, the National Highway Traffic Safety Administration (NHTSA) opened an investigation into whether the vehicle complies with current federal motor vehicle safety standards.

All automakers self-certify that their vehicles conform to these standards, and Tesla has claimed that the Cybercab is compliant. But it is missing features that the current standards require, like a brake pedal and side mirrors. The NHTSA has proposed removing some of these requirements, but that rule change is likely still at least a few months away.

Not every Tesla executive agreed on this during Wednesday’s call.

Asked about the remaining bottlenecks for a wider robotaxi rollout, Ashok Elluswamy, Tesla’s vice president overseeing autonomy, said they are “not purely regulatory.”

“It’s technology-driven,” he said. “We had initially had the robotaxi service from 6 a.m. to 9 p.m., and then expanded to 10 p.m. to 11 p.m, now it’s going to be pretty soon, [to] 2 a.m. So we’re slowly expanding the availability of robotaxi, and that is driven by the capability of the AI software.”

That could be why Tesla’s service still doesn’t operate on highways, and in certain conditions, although it has been spotted operating in heavy rain many times in Austin — something Waymo’s robotaxis have largely avoided. Elluswamy said Wednesday that Tesla plans to tackle more of these challenging driving scenarios by the end of this year, before scaling further.

“We just want to make sure that there’s not a single safety incident. That is our operating principle. And even if there’s the slightest possible risk at all, we don’t like to take on that risk,” he said.

Spotify confirmed the reports of its crackdown on Family Plan sharing during its third-quarter call with investors, calling its new method “a routine improvement” to its existing verification tools.

Over the past several days, reports had surfaced indicating that Spotify was rolling out a “QR code verification” method, where the app would prompt members on its Family Plan to verify they lived with the plan manager by scanning a QR code. This code appears on the plan manager’s phone, which the plan member then needs to scan with their own phone, according to Spotify’s support documentation.

Plan members have seven days to complete the QR code after getting a notification, which means they could be out of luck if they live outside the home and aren’t planning a return visit anytime soon. If the verification window is missed, they won’t be able to join another Family or Duo plan for 12 months, but would instead have to shift to an individual plan that they pay for themselves.

The move is meant to boost subscriptions to Spotify’s profitable Premium plans, rather than allowing adult children or friends who live apart from gaming the system to get a cheaper plan by splitting a Family Plan subscription.

During the earnings call, an investor asked a question about the crackdown and what it means, but Spotify co-CEO Alex Norström downplayed the change, calling it a “routine improvement,” not a “crackdown.”

“Family Plan has existed for a long while. It has been around for, I think it is now close to 12, 13 years in Spotify history, and it makes up a pretty significant proportion of our subscriber base,” said Norström. “As such, it is important for us to just make sure over time that we improve the verification methods that we have. There are certainly ways, over time, to find holes around it. As that happens, we figure out how to improve it over time,” he added.

The executive said that the company’s attempt to better verify its users is not something that would have an end date or would ever be complete, but would be considered an ongoing routine improvement.

In today’s market, no single metric defines what a Series A is, or how to raise one. At this year’s TechCrunch Disrupt, three investors offered founders advice on raising one, and one of them tacked on a warning. With money pouring into AI and revenue numbers being inflated to justify sky-high valuations, a correction could come as soon as 2027.

The warning came from Shailendra Singh, a managing director at Peak XV, who spoke alongside Jahanvi Sardana, a partner at Index Ventures, and Janelle Teng Wade, a partner at Bessemer. “Seed and venture, I’ve never seen it as dysfunctional as it is now,” Singh said, adding that the current market is one of the hardest to raise in that he has seen.

“The burden is high to convince any top firm that you are amongst the few special companies that can break out, and then the dispersion is very high on the terms you might get, how much capital you can raise, and what terms you can raise it,” he continued. In fact, it’s common now to see a company trying to close a round with as many as three different prices at once. “[I]’s a crazy sign of market froth,” he said.

“We’re calling it the ‘tranchopalooza,’” Sardana added, referring to the way rounds are being split into tranches, or separate chunks.

Beyond the warnings, the investors laid out what they look for in a Series A pitch, including, first, a persuasive story about why a founder has a shot at building a big company. “Founders who can attract very high-quality talent have an advantage,” Singh added.

What a founder builds matters, too. Teng Wade said capital is flowing into areas such as defense, robotics, and AI infrastructure. She also noted that foundational model companies are moving toward vertical integration (taking on more of the stack themselves), but there’s still room for newer players, including neo-clouds that rent out AI computing power and neo-labs focused on AI research.

Metrics, meanwhile, still help a lot, but they are less defined than they used to be, especially in an era like AI where it’s incredibly expensive to run a company. Teng Wade gave an example of how her firm evaluated Anthropic in its early days, when its margins were negative, and it was spending heavily on compute.

One partner on her team, however, felt that AI was going to be “bigger than everything we’ve ever expected in terms of price premium on the frontier models,” she said, “becoming a substitute for other types of software.” The logic was that even with money-losing margins, the market would be big enough to pay off if the company became one of the top three players. Anthropic is now reportedly preparing an IPO at a valuation of around $2 trillion. It told investors this month it was profitable on an adjusted basis, though that measure excludes some costs.

The investors also spoke about the importance of honesty in a market like this, where inflated annual recurring revenue (ARR) numbers are swirling alongside headline valuations. “I tell founders, If you ever show a chart which shows cumulative revenue or cumulative ARR, that is a massive red flag,” Singh said. “It means that your time series data is not good because cumulative data is a big lie.” Added Singh: It’s important for a company to be intellectually honest in how they are presenting and engaging in bull cycles.”

Singh said the market is full of “false positives” because “it can’t be that every company is going to be successful. This is as optimistic as I’ve seen the venture world in two decades, and normally it’s a sign. I think 2027 will be different. I think we will see a correction.”

Finally, the panelists were asked to give founders at least one piece of parting advice. Singh went first, reminding founders that building a company is one of the “most excruciatingly difficult things” one can do. And with that, “deal with it,” he said. He also urged founders not to rush into something, to watch out for weak product-market fit, and always hold the bar very high for what they are trying to solve.

Sardana said it is important to know how to pitch but that founders should be careful not to let that dictate what they build. “It feels like [founders are] deciding what to build based on what’s getting funded, which is really scary,” she continued. Teng Wade, meanwhile, said her typical advice of telling founders to be conservative with their projections and requests hasn’t really held up.

“I would say just be audacious in your ask,” she now says, and see how the market responds, “and then you can tailor from there.”

OpenAI made its Decisions API model widely available to users for the first time, entering the market for fast, cheap intelligence that recently made TypeSafe, a startup founded by a former researcher at the frontier lab, a $7.6 billion company.

TypeSafe’s Jev model created a category for deep learning models that act as super classifiers: Given a set of options, they can quickly make choices and offer a confidence rating in their decision. Software developers quickly found that these decision or System One models could replace LLMs in a variety of workflows, operating more quickly and a at dramatically lower cost.

Nikunj Handa, the product lead for OpenAI’s APIs, told TechCrunch that his colleagues saw Jev and said “wow, we we have like a million use cases for this.” Handa’s team sprinted to offer their own version: Rather than building a specifically calibrated decision model, the team took ChatGPT Luna and reconfigured it to speed up the inference and lower the cost.

“The big unlock with Decisions API has been the speed, shrinking things from one to three seconds to 100, 120, 180 milliseconds,” Handa said. “In the future, we may sort of like have specific training to make our models like even better in these ways. But right now, it’s just plain old Luna that gets the job done.”

Announced at the company’s developer day on September 29, the Decisions API became available to all customers on Thursday. OpenAI added new voice and visual modalities to the public version, and worked to ensure lower latency in regions around the world by targeting its inference resources geographically.

In initial trials with customers like Box, Datadog, Ramp, and Speak, engineers used the API to replace OpenAI’s LLMs with higher throughput. Andrew Hsu, the CTO at Speak, which builds AI language tutors, said the Decisions API allowed the company to cut the time between actions from 1.4 seconds to .13 seconds. OpenAI also used the product to improve their Dots voice assistant’s responsiveness.

Still, it’s not clear exactly what makes a good decision model, other than “it works for you” — optimization across intelligence, cost, calibration, and speed seems to differ by workload. TypeSafe CEO Diogo Almeida distrusts benchmarks and urges people to trial the models in their own work, but even TypeSafe released its own rankings last week alongside the release of its latest Jev model, which includes new features including support for images and smart-if statements.

Public benchmarks like the AIM System One Index and the Artificial Analysis Decision Index tend to favor D1 Ultra, a new decision model that Liquid released last week, and Base Lab’s Elephant-1. The latest Jev model remains competitive on both indices, in seventh and third, with OpenAI’s Decision Model falling eighth and fourteenth, respectively. However, TypeSafe’s offerings are, for now, the most widely-queried on OpenRouter.

The most fun Jev benchmark, however, is JevMan, a benchmark where decision models play PacMan; the latest Jev model tops that ranking. Felix Wunderlich, the chief growth officer at Opper AI, which built the benchmark, said the classic game “is an interesting test because of its nonstop stream of quick, high-stakes decisions. At every junction the model must decide where to go in a few hundred milliseconds.”

Where OpenAI’s arguably more capable model may have an advantage is in multi-modality—for image classification, “Luna is extremely powerful and has been showing better calibrations,” Handa told TechCrunch, adding that OpenAI customers want the Decisions API to be more flexible about using images as outputs as well.

Handa also saw further increases in speed enabling new use cases: “Can we get this to something like you know 60 milliseconds, and then you could imagine like an API like call like for every single frame in like a video?”

As these offerings get faster and cheaper, we can expect them to live up to Jev’s namesake, an economist who explained how using a resource more efficiently makes it more ubiquitous. Or, as Handa sees it, there will be an expanding “crazy set of use cases” that were hiding in plain sight until the decision models revealed them.

10月20日甲骨文發布2026年第四季度的關鍵修補更新，總共處理1,209個漏洞，沒想到隔天又發布新的資安公告，主角是CVSS 3.1嚴重度評分為9.1分的CVE-2026-91403，警示此漏洞可透過網路利用，無須使用者互動，而且攻擊的複雜度低，目前已知受影響的產品都與Java平臺有關，包括GraalVM Enterprise Edition（21.3.20版）、GraalVM for JDK（17.0.21版與 21.0.13版），以及Java SE（8u511版、8u511-perf版、11.0.33版、17.0.21版、21.0.13版、25.0.5版、27.0.1版），漏洞存在的元件是Java安全通訊端擴充套件（Java Secure Socket Extension，JSSE）。值得注意的是，目前Oracle的這份公告尚未直接列出已修補版本，而「Patch Availability Document」的超連結則須通過甲骨文技術支援網站的身分驗證才能存取。

關於漏洞利用方式與發生位置，甲骨文表示，攻擊者可透過向API提供資料的web service的API，利用CVE-2026-91403；此漏洞也適用於 Java 部署環境，通常發生在執行沙箱化 Java Web Start 應用程式或沙箱化Java applet的客戶端中，而且，這些應用程式會載入並執行不可信的程式碼（例如來自網際網路的程式碼），並仰賴 Java 沙箱來確保安全性。

關於漏洞成因，甲骨文的資安公告並未交代，不過，根據資安新聞網站Cyber Kendra的報導，與CVE-2026-91403資安公告同樣是10月21日發布的OpenJDK公告指出，此漏洞與8個遭外洩的中介CA憑證有關，也就是10月20日曝光的印度數位憑證業者eMudhra中介CA憑證外洩事故，而Java正是受到此事影響的程式執行環境之一。