AI

OpenAI’s safety systems are supposed to watch the AI. On two occasions this month, the AI turned them off instead. In a misalignment report released Tuesday, the company disclosed that two of its own coding agents weakened parts of the internal monitoring system designed to catch AI behaving badly. On October 3, an agent running on GPT-6.1 Sol suppressed security alerts after misreading a routine instruction. Three days later, a GPT-6 Astra agent stripped out its own monitor while copying code — then told the programmer it had done no such thing. ...

Microsoft’s AI team taught a machine to practice medicine the slow way, and the results are hard to ignore. In a study published Thursday in the journal Science, Microsoft researchers reported that their MAI-DxO diagnostic system correctly diagnosed 304 fiendishly complex medical cases 83% of the time. General physicians given the same cases managed 36% with textbooks and the internet, and just 22% without references. That’s better than a two-to-one edge for the machine. ...

The AI copyright wars just had their biggest week yet, and a music publisher you’ve probably never heard of fired the loudest shots. Reservoir Media filed twin lawsuits on October 21 against Suno and Anthropic, accusing both companies of building their AI businesses on 1,903 songs they never paid for. The combined damages sought: nearly $340 million. Both cases landed in the same courthouse — the US District Court for the Northern District of California — and both were filed by the same lawyer, Richard S. Busch of Adams & Reese. Busch is the litigator who won the famous “Blurred Lines” copyright case for Marvin Gaye’s estate, and he has been on a tear this month: his firm already filed near-identical suits for Round Hill Music against Suno and Anthropic, plus one for Gerencia 360 against Suno. ...

OpenAI just picked a fight with a $7.6 billion startup by copying its best trick. On Thursday, the company opened its Decisions API to all customers, a new endpoint that doesn’t write paragraphs or hold conversations. It makes split-second structured calls: is this product photo damaged, should this ticket get escalated, which option should the agent pick next. The price starts at $0.10 per million input tokens, with output free. ...

Jensen Huang says Nvidia’s era of writing giant private checks to AI labs is coming to an end. Speaking on Wednesday at the Morgan Stanley Technology, Media & Telecom conference in San Francisco, the Nvidia CEO said the company’s $30 billion investment in OpenAI “might be the last time it has the opportunity to ‘invest in a consequential company like this.’” The $100 billion deal the two companies discussed in September 2025, he added, “does not seem feasible.” ...

The three companies racing to build the future spent the third quarter of 2026 buying influence over how that future gets regulated — and each one set a new record for lobbying spending doing it. Nvidia, Anthropic and OpenAI poured a combined $7.5 million into federal lobbying between July and September, according to an analysis of new disclosure filings by POGO Investigates, with each company posting its highest quarterly total ever. ...

Microsoft put a name on the ConnectWise hack Wednesday, and it’s a grim milestone for defenders. The company said Storm-0940, a China state-aligned hacking crew, was behind the recent supply-chain compromise of the ConnectWise Cloud Backup application — and ran its follow-on attacks through ARTEX, an agentic AI penetration-testing framework that did the break-in work at machine speed. The numbers are the story. Since September, Storm-0940 targeted hundreds of organizations worldwide, with attack attempts touching more than 2,250 Microsoft tenants at peaks of 2,900 sign-in attempts per minute. In several cases, the time from first targeting to privileged access on a victim’s cloud infrastructure was under three hours. Nobody had to type a password. ...

Meta’s earnings beat didn’t matter. Shares of Meta Platforms dropped about 3% in premarket trading Thursday, October 22, 2026, as investors weighed a triple hit: a stunning $25 billion penalty ordered by a New Mexico judge, a plan to issue non-voting shares, and a capital spending forecast of $135–145 billion to fund the company’s AI buildout. Meta was the first of the AI hyperscalers to report this quarter, and Wall Street didn’t like what it saw in the fine print. ...

OpenAI is turning up the ad dial. The company said Wednesday it is switching on ChatGPT ads in 26 more countries, taking the total to 89 and revealed that more than 100,000 advertisers have now run campaigns since testing began in the US earlier this year. The expansion lands as OpenAI races to prove it can make serious money from the hundreds of millions of people who use ChatGPT for free. ...

Anthropic has its own answer to the AI agent wave, and it arrived Wednesday. The company opened a research preview of Claude Duet, a standalone app that turns Claude into a proactive agent aimed at professionals and small-business owners. “It comes to you with what needs your attention, completes tasks you ask of it, keeps working even when you step away, and checks with you before it acts on your behalf,” Anthropic said of the new app, as reported by Engadget. Access runs through a waitlist, starting with Pro and Max subscribers in the US. ...

OpenAI is putting its AI models on Google Cloud. The company announced Wednesday that its frontier models — GPT-Astra, GPT-6.1 Sol, and GPT-6 Luna — plus its Codex coding agent will land in Google Cloud’s Model Garden and Marketplace. General availability is expected in the coming weeks, though select enterprise customers are getting private-preview access starting today. What’s actually changing Google Cloud customers will be able to buy OpenAI models directly through the Marketplace and count the spend against their existing cloud commitments. No separate procurement dance, no second billing relationship. The models deploy from Model Garden, Google’s catalog for third-party and first-party models, and all usage lands on a single Google Cloud invoice that also draws down Flexible Savings Plans. ...

ChatGPT finally has a way to hide conversations without deleting them. OpenAI began rolling out Locked Chats on Tuesday, October 20, a feature that lets users put a four-digit PIN, or Face ID / Touch ID on supported devices, on individual conversations, keeping them out of regular chat history, pinned chats, and search results. The scenario OpenAI has in mind is painfully ordinary. You’re handing your phone to a friend to show them something, and the last thing you want is a personal conversation sitting there in the sidebar. With the lock on, that thread simply doesn’t appear until you authenticate. Tap a locked chat and it moves into a dedicated “Locked Chats” section, kept apart from everything else. ...

Six months of construction, fourteen days of hookup. That was the pitch Nvidia CEO Jensen Huang brought to Berlin on Wednesday, and it amounts to the company’s most aggressive move yet against the biggest bottleneck in the AI buildout: the time it takes to actually turn a site into a working data center. The vehicle is Nvidia DSX POD, unveiled at the GTC Berlin keynote on October 21. It’s a reference design for complete Vera Rubin AI factories that partners build inside modular containers. Each container carries computing, networking, cabling, rack-level power, and liquid cooling, all assembled, integrated, and validated in a factory before the unit ever ships. Once it lands, a prepared site just connects power, cooling, and network, and the whole thing can go live in as little as 14 days, down from the six months a traditional onsite build takes. ...

Anthropic lost $80.7 billion in six months. On paper, it’s one of the largest corporate losses ever recorded, and the company plans to go public next month anyway. The Wall Street Journal reported Tuesday that it had reviewed a draft of Anthropic’s S-1 prospectus covering the first half of 2026. The filing offers the most honest look yet at the Claude maker, and the numbers are staggering: $16.3 billion in revenue, up from $1.1 billion in the same period last year, sitting alongside a net loss of more than $80 billion. That’s about $13.5 billion lost every single month. Anthropic declined to comment on the Journal’s report. ...

OpenAI wants $30 billion, and it wants the money before Anthropic starts selling itself to the public. Reuters reported Monday that the ChatGPT maker is racing to lock in investor commitments for a funding round of about $30 billion that could value it at more than $1.4 trillion — with the timing anything but subtle. The company hopes to close the round before rival Anthropic’s expected IPO in mid-November, when Anthropic’s own financial disclosures would hand investors a side-by-side comparison that neither lab seems eager to face. ...

Boston-based AI startup ASI said Tuesday it has raised $325 million in a Series C round that values the company at $3 billion. The round, led by Spark Capital, is the latest sign that investors are paying up for AI that manages the physical world, not just the chat window. Spark Capital led the financing, with participation from affiliates of Citadel Securities, Meritech Capital, Craft Ventures, Energy Impact Partners and Andreessen Horowitz (a16z), according to the company’s announcement. ASI said it would use the proceeds to expand deployments across energy, logistics and military environments and to add to its research and engineering teams in Boston. ...

One hacker broke into more than 100 company websites. He barely lifted a finger. The AI agents did almost everything. Over the past couple of weeks, a Chinese-speaking hacker compromised websites belonging to more than 100 companies, including farm equipment maker Deere, appliance company SharkNinja and engine manufacturer Cummins, according to an investigation Reuters published on October 19. Each successful break-in cost the attacker about $50 in AI tokens. The human’s role in the whole operation? He wrote fewer than 100 Chinese characters of instructions, then walked away. ...

OpenAI is building a humanoid robot. That’s not speculation anymore; it’s a commitment Sam Altman made in two words on the Sources podcast in early September: “We will.” The quote, first reported by Forbes, came when Altman was asked whether OpenAI’s hardware push would include a humanoid or a data-center machine: “We will definitely do a humanoid,” he said. “We will do other form factors as well.” Asked to pick between the two, his answer was simply: “Both.” ...

Tenstorrent, the chip startup trying to chip away at Nvidia’s grip on AI hardware, raised $1.5 billion on Monday at a $6.2 billion valuation. The round more than doubled the company’s value from the roughly $2.6 billion it fetched in late 2024. That makes the 10-year-old startup one of the best-funded Nvidia challengers on the planet. CEO Jim Keller told Bloomberg the company turned down acquisition interest from Intel and Qualcomm, reportedly at valuations of up to $5 billion, to stay independent and raise more capital instead. The deciding factor, Keller said, was Tenstorrent’s Galaxy servers: the company showed they could run DeepSeek’s open-weight models, and decided the business was worth building, not selling. ...

The federal government just told DeepSeek to get off its computers. The Department of Homeland Security is ordering the removal of DeepSeek products (and any third-party services built on the Chinese AI model) from US civilian agency IT systems, according to a notice published Monday, Oct. 19. It’s also banning future purchases. Civilian agencies can’t buy DeepSeek products or acquire third-party tools developed using DeepSeek going forward. DHS’s reasoning was blunt: the models pose “multiple threat vectors,” and their use “may result in the transfer of sensitive data to the PRC if used by US government employees or vendors.” ...