最新 AI 资讯
OpenAI Apologises for AI Breach in Australia
This comes after PM Anthony Albanese subsequently expressed “extreme concern” to OpenAI CEO Sam Altman and criticised the company for taking “way too long” to notify Australian authorities.
View
Anthropic Launches Claude Sonnet 5.5 With 30% Lower Costs Per Task
The new model costs $2 per million input tokens and $10 per million output tokens, while Anthropic says it requires fewer tokens to complete the same tasks.
View
Anthropic’s prospectus details losses, growth, and, yes, a warning that its AI could end humanity
Anthropic devoted nearly a third of its hotly anticipated IPO prospectus to risk factors,according to the Financial Times, which says it has reviewed the filing in recent days. The filing details specific, worrisome behaviors that Anthropic says its models have already shown or could show, including attempts to “resist shutdown,” to “conceal or manipulate information,” and behavior “resembling blackmail,”according to Reuters. The disclosures are decidedly grim for a company whose own backers believe it could list above $2 trillion, more than double its $965 billion valuation from May, in potentially the biggest IPO ever. It’s a strange position for any company to be in — warning that its product could end humanity, while making some of its earliest investors and employees extraordinarily wealthy in the process. Reuters wasfirst to reporton the financial details within the prospectus on Monday, saying Anthropic recorded an operating loss of more than $8 billion in 2025 as spending on computing power surged, and that its revenue jumped twelvefold to nearly $4.6 billion, though rising infrastructure costs last year pushed total operating expenses to almost $13 billion. Also per Reuters, Anthropic’s prospectus further reveals plans to spend a whopping $518 billion on cloud, computing and infrastructure in the coming years. (Anthropic has already inked compute deals this year withGoogle,SpaceX, andNscale, among others toward that end.) The FT meanwhile reports that Anthropic’s numbers have moved even faster in 2026. Its second-quarter revenue alone reached $11.5 billion, and the company is on track for its second straight quarter of operating profit on an adjusted basis. According to the FT, the prospectus also flagged customer concentration, with nearly a quarter of last year’s revenue coming from just two clients. (No word yet on who these are.) The disclosures, which reportedly include “existential risks to humanity” — a first, judging by a quick scan of the SEC’s database — comes as hand-wringing quickly grows over AI safety. CEO Dario Amodei has spent the month publicly calling to “pace the frontier” of AI development, telling the UN Security Council last week that AI couldthreaten humankindand calling it “the most important global security issue facing the world today.” Rivals Sam Altman and Elon Musk have backed him up, too, in a rare moment of solidarity for competitors who’ve seemingly relished opportunities to disparage each other publicly. Another rival, Mark Zuckerberg, has meanwhileswatted away concerns, telling NBC News last week that he doesn’t “think that we need some kind of industrywide coordination.” The warnings follow a string of security incidents in which AI agents have breached outside systems. In fact, OpenAI disclosed last week that its tools have hacked “dozens” of external sites, including government one, includingthe SEC’s site itself. Earlier on Monday, it said it hadscrapped plansto release its newest model owing to safety concerns.
View
Russia’s Sber Launches GigaChat 3.5 Reasoning to Take on Complex AI Tasks
The Russian bank said GigaChat 3.5 Reasoning is the only Russian open-source model with a reasoning mode.
View
AMD Acquires World Labs for $8.2 Bn, Fei-Fei Li to Join as Chief Scientist
The deal brings World Labs’ spatial intelligence research into AMD, with the companies working towards an open AI ecosystem.
View
AI Needs $6 Tn Annual Revenue by 2031 to Justify Compute Spending: Bain Report
Bain expects new AI products across autonomous systems, robotics, physical AI, and other emerging applications to drive $4.2 trillion in additional revenue.
View
The AI boom took over Climate Week and not everyone is happy about it
It was the best of times, it was the worst of times … I’ll spare you the rest, but the cliché Dickens line really does sum up this year’s New York Climate Week. Much of the climate tech community — like the rest of the U.S. economy — is eagerly riding the AI wave. Some have reservations about thesheer quantity of natural gas power plantsbeing built to power AI data centers. But because many climate tech startups are energy-focused or energy-adjacent, the buildout has been embraced as an opportunity to get companies through thevalley of death. The singular focus also means some promising sectors risk being overlooked. It’s a continuation of a trend that’s emerged over the last year. As climate tech companies struggled to get financing — either because of canceled federal grants or investor hesitancy — those that could change their pitch to match the AI mania did so. The pivot has helped many climate tech startups land fresh funding from investors. Total venture deal value has risen for four consecutive quarters, cresting the $14 billion mark in the first quarter of this year, according to the most recent available data from PitchBook. It’s the best fundraising environment for climate tech in the last few years, with most of the deal value driven by sectors boosted by data center construction, including the built environment, grid infrastructure, and dispatchable energy that can be turned on or off when it’s needed. It’s an opportunity few have wanted to pass up. One exchange during a panel at New York Climate Week captured the moment: Two founders, when asked whether they’d prefer the AI buildout to proceed at its current pace or at a more climate-responsible speed, said without hesitation that faster was better. Unsurprisingly, both of their startups were in energy. And yet not everyone agrees. I heard from several founders who felt that the data center boom was distracting from other promising segments of climate tech, including those that were meeting their targets without having to rely on AI mania. “Corporates are still interested in climate,” one founder told me. The difference today is that large companies don’t want to crow about it, mostly for fear of drawing the Trump administration’s ire. There were also signs that the AI boom was beginning to wear thin on some. For many startups, money for scaling was hard to find three years ago, even if they were showing promising results. Now, customers are clawing their way into demos. “Where was this money three years ago?” I asked several people. I received more than a few knowing eye rolls in reply. It’s the world they live in these days, they acknowledged. The smart entrepreneurs are all finding ways to meet customers where they are. Ultimately, the undercurrent at New York Climate Week was that the data center party won’t last forever, but it might last long enough to help startups build durable businesses. Once that happens, they can refocus on the carbon-cutting mission they were founded to pursue.
View
Source: Inference provider Modal Labs closing in on $750M round at $15.75B valuation
AI inference infrastructure provider Modal Labs is nearing a $750 million funding round led by Accel at a $15.75 billion valuation that includes the investment, according to a source with knowledge of the funding. The size of the round has not been previously reported, though Axios and Bloomberg have reported other details of the deal. The new round would more than triple Modal’s valuation from the $4.65 billion it reached when it announced its $355 million previous fundraise just four months ago. Modal Labs declined to comment. The deal comes amid soaring demand for inference services, the process of running an AI model that’s already been trained to generate outputs, particularly from customers relying on open-source models. Other inference startups are also in talks to raise fresh capital at much higher valuations. Baseten is nearing an infusion of capital at a $26 billion valuation, doubling what it was worth in June,Bloomberg reported. Meanwhile, Fireworks and Fal, a startup providing inference for video and image generation, have also talked to investors about new rounds that would significantly increase their valuations, according toThe Information. Although revenue for these companies has been growing rapidly, their margins are thin, largely because the cost of acquiring or leasing compute remains very high. Fireworks announced in July that its annualized revenue had hit$1 billion, a fivefold increase from the year before. Multiple inference-focused startups are expected to reach the same revenue milestone by year’s end, according to our source. Modal was founded in 2021 by CEO Erik Bernhardsson and CTO Akshat Bubna. Bernhardsson, who is Swedish, spent more than 15 years building data teams at companies including Spotify, where he helped build the music-streaming service’s recommendation system, and Better.com, the online mortgage lender, where he served as chief technology officer. Bubna studied math and computer science at MIT and was an early staff engineer at Scale AI, the data-labeling startup, before co-founding Modal. The company, which is based in New York and estimated to have roughly 150 employees, lets developers train AI models and run other compute-heavy workloads without managing their own servers. Its web page lists customers that include the coding startup Cognition, the AI music generator Suno, the fintech company Ramp, and the publishing platform Substack. As of May, Modal had surpassed$300 millionin annualized revenue, it told Reuters at the time. The fundraising talks come two months after Modal was pulled into one of the AI industry’s most closely watched security incidents. In late July, Modal disclosed that a customer’s data had been compromised as part of the same hacking campaign carried out by a rogue OpenAI agent against Hugging Face. Modal Chief Technology Officer Akshat Bubna said the breach traced back to a flaw in a customer’s own code, not to Modal’s systems. “We’re aware a Modal customer published an unauthenticated endpoint that allowed anyone on the internet to use their sandboxes for code execution,” Bubnasaid in a statementto press outlets at the time. “This was used by the rogue agent. Modal’s platform was not compromised in any way,” he’d added.
View
OpenAI reportedly ditches model over safety concerns
OpenAI had planned to release yet another AI model next month, but has decided to nix the release over safety concerns. The Wall Street Journalreportsthat Astra 6.1 was scheduled to be released as soon as within the next few days. However, the model “showed higher levels of deception” than previous models and exhibited unsafe behavior, the Journal writes. Saachi Jain, OpenAI’s head of safety systems, told the WSJ that the model tested poorly on alignment, a measure of how well the program adheres to human intent. TechCrunch reached out to OpenAI for more information and will update the article if it responds. Astrawas releasedearlier this month and hailed by OpenAI as its most powerful model yet. Questions about safety have plagued the AI industry over the past several months — ever sincethe Hugging Face incident, in which an OpenAI agent broke free of its sandboxed environment and hacked several different companies. Since that incident, more models — includingAnthropic’s ClaudeandGoogle’s Gemini— have been revealed to have exhibited similar behavior. The deluge of concerning stories has, ironically, helped to push the policy conversation in the U.S. toward an outcomedesired by top AI labs: the institution of new industry standards for AI safety and potentially aslowdownof the industry itself. Companies like OpenAI and Anthropic have claimed that the concern here is safety, although another potential motivation posited by critics is that it couldentrench the industry positionof those companies at the detriment of less resourced firms.
View
Peak XV ups Surge seed investment ceiling to $5M, unveils 18-startup cohort
Peak XV Partners, one of the largest venture capital firms investing in markets including India and Southeast Asia with more than $10 billion in assets under management, has increased how much it invests per startup through Surge, its seed-stage investing platform, as it unveils a new cohort of 18 companies. At least three of the companies in this cohort had already raised outside funding, in some cases from Peak XV itself, before joining Surge. The new batch, called Surge 12, is the first to operate under Peak XV’s higher investment ceiling of up to $5 million per company,up from $3 million previously. The venture firm invested more than $50 million across the cohort, which has collectively raised over $90 million in seed funding, according to Peak XV. Its median investment per company has also increased, though the firm declined to disclose the figure. “The bar to raise a Series A has gone up pretty significantly,” Rajan Anandan (pictured above), managing director at Peak XV, said in an interview. He added that the firm is also seeing more capital-intensive companies, particularly in deeptech, that are raising larger rounds at the seed stage. Surge has become more global with each cohort, Anandan told TechCrunch, with its latest group spanning founders and companies from San Francisco to Sydney. Just five of the 18 startups in Surge 12 are focused on the Indian market, while more than half of the companies are based in India. The remaining 13 target global markets, highlighting the difference between where the companies are built and where they expect to find customers. Since itslaunchin 2019, when Peak XV operated as Sequoia Capital India and Southeast Asia, Surge has backed more than 180 startups founded by entrepreneurs representing more than 18 nationalities. Peak XV says the 10 largest companies to emerge from those cohorts now generate more than $1 billion in combined annual revenue. Anandan described Surge as one way Peak XV invests at the seed stage, alongside its standard seed investing, while the firm still remains an investor as companies progress through later funding rounds. The founders it backs typically include repeat entrepreneurs, experienced operators, and highly specialized technical founders, he said, with about 50% to 60% of a typical cohort made up of people coming from operating roles at established technology companies. This cohort’s startups span AI, robotics, space, consumer products, healthcare, music, and fintech, ranging from AI safety and personal computing to autonomous robots built for underground pipes and satellites designed to detect radio-frequency signals from orbit. Alma— founded by Nischith Shadagopan M N and Vinod Ganesan — is building a personal computing platform focused on making computer use faster and more affordable. Its founders previously worked at Microsoft Research and were founding engineers atSarvam AI, a Bengaluru-based startup building AI models for Indian languages. August AI— founded by Anuruddh Mishra, an IIT-BHU alumnus who started the company in 2022 after a personal medical misdiagnosis — provides a healthcare platform that combines AI with physician-led care, reaching over 9 million users across 160 countries. Ditto— founded by UC Berkeley dropouts Allen Wang and Eric Liu — works as an AI dating matchmaker inside iMessage, aimed at helping college students turn digital introductions into in-person connections. (TechCrunch wrote more about this onelast month.) The company had already raised $9.2 million in a Peak XV-led seed round announced earlier this year. GameStock— founded by Antoine Mistico, Easton Dana, and Vivek Indlebele Narasimha Prasad — brings competition mechanics to financial markets, turning investing and trading into a more competitive experience. Mistico is a two-time founder and former professional baseball player. HiLoop— founded by Jad Ghalayini, Karan Brar, and Thomas Boser — helps AI companies adapt general-purpose open-weight models for specific applications using its post-training platform. Its founding team includes former Reducto engineers and a Cambridge computer science PhD who completed his doctorate at 24. Hoola Health— founded by Deeksha Senguttuva — focuses on care for children and their families, providing consultations, vaccinations, medicines, diagnostics, developmental therapy, and dental services on a single platform. Senguttuvan grew up around healthcare, as her family built and operated a hospital group. Kello— founded by Mona Gandhi and Subramanya Jingade — is building an AI-powered talent-discovery platform focused on identifying a candidate’s potential and trajectory rather than relying primarily on conventional credentials. Gandhi says she was Airbnb’s first female engineer and she previously founded Upraised, while Jingade previously co-founded AmbitionBox. Kindling— founded by Adam Miller and Sachin Shah — is building what it calls a “storytelling operating system” for technology startups, using AI to help companies develop and produce their communications and content. Puralink— founded by Harrison Crowe-Maxwell, Shyeon Delnawaz, and Thien “Long” Tran — is developing autonomous robots that can navigate underground pipe networks. Crowe-Maxwell has been building robots since childhood and turned university research into the patented drive technology behind the startup. Reinforce Labs— founded by Anish Das Sarma — is developing tools to evaluate, red-team, and remediate enterprise AI systems. Sarma previously founded a company acquired by Airbnb and later served as a director at Google, where he led AI and machine-learning teams. Riffle— founded by Anurag Choudhary and deo — is building a browser-based platform where musicians can create, collaborate on, and share music, reducing the need to move between separate tools during the creative process. Rosella— founded by Chris Dwyer and Sean Stuart — is building an AI-native commercial insurance brokerage for U.S. businesses, using AI to automate parts of the traditionally manual process of finding and placing business insurance. Rosella raised a roughly $2.5 million pre-seed round led by Peak XV and Intact Private Capital earlier this year. Tribe Money— founded by Himanshu Arora and Nikhil Shanker — gives an AI-powered personal finance platform that helps users track their money, research investments and make investing decisions. ULOOK— founded by Adheesh Boratkar and Siddhesh Ravindra Naik — is building autonomous satellite systems for radio-frequency sensing and spectrum intelligence, targeting customers globally. Its founders have worked on more than 12 satellite missions. The company had already raised roughly $2.3 million in seed funding from growX Ventures and InfoEdge Ventures before joining Surge. Wingit— founded by Nikunj Kothari and Saksham Khandelwal — is building a beauty platform aimed at India’s growing premium-consumer market. It is focused on how consumers discover and shop for higher-end beauty products. Three other startups in the cohort have yet to publicly reveal their names or products. Peak XV said they are working in education, applied AI, and medical products.
View
Google is killing off Gemini’s Gems in favor of ‘skills’
As all-in-one AI agents like Meta’sMuseandInstincttake off, Google announced it’s shutting down the Gemini feature known as “Gems,” which had allowed users to build custom AI assistants for specific tasks. However, the work users invested in creating the Gems won’t be destroyed. Gems will be automatically migrated to “skills” that can be used across different AI tasks. Details about the change are being shared in the Gemini app, where a message warns users that Gems will become skills starting on November 17, 2026. The company said it will migrate the Gems to the new format, so users won’t have to do anything to make the transition. The Gems themselves will remain usable until then. Launchedin 2024, Gems were meant to helpusers teach their AI to perform certain taskswithout having to repeat the instructions. For instance, some of Google’s pre-made Gems had included a learning coach, a brainstorming assistant, a career guide, a coding partner, and an editor. Users could also make Gems for their own needs, like a running coach, nutritionist, or vacation planner. These custom assistants could also beshared with others, which Google had hoped would help make its Gemini AI app more popular. The news of Gems’ shutdown is another example of why Google shouldn’t be so quick to give every new AI feature its own brand name, icon, and prominent placement in its app’s navigation — especially if it’s going to shuffle things around over time, merging one feature into another. (To be clear, this has been a failing point of Google’s strategy long before the AI era. At one point, for instance, the company was operating multiple different messaging and communication apps at the same time.) Yet, even as skills, the former Gems still aren’t as consumer-friendly as just typing in text to a chatbot like Meta’s Muse. Instead, Google notes you’ll have to enter a forward slash “/” in a task thread to select the skill you want to use — a user interface that engineers, not regular folks, tend to prefer. Gems’ wind-down was first reported over the weekend by9to5Google.
View
Anthropic releases Sonnet 5.5, which it calls a significantly cheaper, faster work partner
As the AI model wars continue, Anthropic has released the newest version of Sonnet, the company’s mid-tier model, which it says will work much faster (and for significantly less) than its predecessor. The lab describes Sonnet 5.5 as an ideal assistant for everyday tasks — including coding and creating office documents. 5.5’s predecessor, Sonnet 5,was announcedabout three months ago. At the time, the model’s selling point was efficient agentic deployment — the ability to run agents at a lower cost than competitors. The big selling point with 5.5, meanwhile, is speed. Anthropic claims that Sonnet 5.5 is 30 percent faster than its predecessor, and that its rate of token burn is significantly slower. In the Anthropic hierarchy of models, Sonnet is less powerful than the Opus model, but can be more useful in certain circumstances because its agility. In particular, Anthropic’s benchmarks show Sonnet 5.5 performing better than Opus 5.5 on agentic coding, likely because of its ability to spawn multiple agents without exceeding cost limits. Sonnet 5.5 is also said to have significant cyber capabilities, with the company claiming that it has “comparable” cyber capabilities to Opus 5. As a result, Anthropic says that 5.5 is the first Sonnet model that will be subject the same cyber safeguards that apply to Fable and Opus. The company also plans to release a new version of Haiku — its smallest model — in the coming weeks, although it didn’t give a firm date as to when that would happen. The last year has seen a flurry of new model releases from the major AI labs. Just last week, OpenAIreleased a number of new models— including enhanced versions of Sol and Luna, its mid-tier and budget-friendly models. Meta alsoannounced a new model, which it said would power an upcoming feature associated with its smart glasses.
View
