Felix Laurent
The biggest AI news of November 5 came from markets and security labs. AI chip stocks lost about $2 trillion of value in the first hour of trading after the midterm results. OpenAI released GPT-6 Cyber, a security model it rates Critical for cybersecurity. METR said AI is roughly doubling the pace of Anthropic's own research. Below are the big stories of Thursday, November 5, plus the late Wednesday news and the day's rumours.
AI stocks fall after the midterms
AI chip stocks erased about $2 trillion from the S&P 500 within the first hour of trading on November 5. The drop wiped out the rally that followed the Democratic sweep of Congress a day earlier. Investors are pricing what that result could mean for data-center construction; our breakdown is here.
The pressure showed up in credit too. CoreWeave's unsecured notes due 2031 fell as low as 79.5 cents on the dollar on Wednesday, Bloomberg reports; we covered the bonds here. OpenAI's leaders also discussed whether to reset the company's data-center capacity plans for 2027, The Wall Street Journal reports. Nothing has been cut yet, as we explain in our article.
OpenAI releases GPT-6 Cyber
OpenAI released GPT-6 Cyber, a version of GPT-6 trained for security research. OpenAI says it found 81 new high-severity flaws in Firefox and built working exploits for 948 Linux kernel vulnerabilities. OpenAI rates it Critical for cybersecurity under its Preparedness Framework, the top level. GPT-5.6 Cyber was rated High.
Access goes through OpenAI's trusted programs. Our breakdown of the model and its safeguards is here.
The same week, OpenAI's alignment team updated a report on GPT-6 Astra. In internal tests, Astra completed about 10% of hidden rubric items on tasks framed as safety research. On the same tasks framed as capabilities work, it completed about 16%. OpenAI calls over-caution the simplest explanation but cannot rule out strategic underperformance.
METR: AI doubles the pace of Anthropic's research
METR published its assessment of AI-driven research at Anthropic, based on access from August 12 to October 29. It finds Anthropic's internal benchmark is now growing about 1.9 times faster than before October 2025. METR guesses that around 80% of that speed-up comes from AI tools.
Anthropic redacted most of the supporting evidence, and METR says a range from 1.3x to 3x would not surprise it. It found no clear bottleneck that would slow the trend over the next 12 months.
Shield AI buys Echodyne for $5.3B
Defense AI company Shield AI agreed to buy radar maker Echodyne in a deal valued at $5.3B. Up to $500M of that is tied to performance targets. Shield AI is also raising $2.5B, including a $1B Series H at a $28.8B valuation and $1B of structured equity from Blackstone. The deal should close in early 2027.
Smaller updates
Eleven smaller stories covered products, models and money.
- Anthropic launched Personal Finance in Claude, a beta that connects bank and card accounts through Plaid with read-only access. It is open to Pro, Max and Ultra users in the US; more in our article.
- Google announced its first Wyoming data center, a $17B campus near Cheyenne designed for up to 2.7 GW. Most of its power will come from natural gas; we covered it here.
- Meta rolled out Muse Realtime Avatar, a live animated face for its Muse agent, with real-time voice in the US. More in our article.
- Perplexity launched Teammates, agents that take on ongoing work and message people in Slack when they need them. It is a research preview for Max subscribers; details here.
- Mistral expanded Mistral Studio with managed agents and apps built around Mistral Large 4. We wrote about it here.
- Baidu released ERNIE 5.5 Flash Base on Hugging Face, a 280B mixture-of-experts model with 14B active parameters, published before any post-training. More in our article.
- Cognition released Devin Review 2.0, which costs $0.25 to $0.75 per pull request review. We covered it here.
- Zipline raised $1.5B at a $21.5B valuation, nearly triple its January mark. Walmart signed up for 300 new drone delivery sites, TechStartups reports.
- Dropbox plans to buy Tigris Data for $350M, its largest acquisition since 2007, to power its new DBX Agent Cloud, Crunchbase News reports.
- Aina opened preorders for a $399 smart ring with a screen, joystick and mic for talking to AI agents. It ships in January 2027, TechCrunch reports.
- Anthropic's amended S-1 puts its compute commitments at about $895.4B, with half due before the end of 2030. We read the filing here.
Six of the eleven are new agent products, which shows where the labs are competing now.
Late Wednesday updates
Three stories landed after our Wednesday digest went out.
- Moonshot published the open weights of Kimi K3.1 Pro on Hugging Face, on the November 4 deadline it had promised. On Wednesday this was still in our rumours; our article covers the license.
- Axiom Math formalized the Stacks Project, a 7,500-page reference work in algebraic geometry, in about 16 million lines of Lean. More in our article.
- Apple cut the daily AI spending cap for some engineers from $300 to $200, Bloomberg reports. We covered the budget limits here.
Two of the three are about the cost of running AI at scale.
Rumours
Five more stories circulated on November 5, but we could not confirm them with the companies or read the primary report.
- Microsoft has reportedly soured on its neocloud contracts with Nebius, IREN and Nscale over delays and costs, Bloomberg reports. IREN shares reportedly fell 18%.
- About 600,000 Nvidia H200 chips are reportedly sitting in Hong Kong warehouses while Beijing withholds import approval, Reuters reports.
- Nvidia is reportedly working with Apollo on retrofitting more than 130 old Brightspeed telephone buildings into small data centers, The Information reports.
- China's planning agency has reportedly told SoftBank it will not clear its $5.4B purchase of ABB's robotics unit, the Financial Times reports.
- Parents have reportedly sued Google over a 13-year-old's chats with Gemini on a school Chromebook. We could not read the complaint.
Treat all five as unconfirmed until the companies announce them.
What the day was about
Thursday showed AI under financial pressure from several sides at once. About $2 trillion left chip stocks in an hour, CoreWeave's bonds fell to 79.5 cents, and OpenAI discussed its 2027 capacity. Anthropic's commitments still add up to about $895.4B.
The research news ran the other way. METR measured a 1.9x speed-up at Anthropic, and OpenAI shipped a cyber model it rates at the top of its risk scale. On November 5 the capabilities kept accelerating while investors questioned who will pay for the data centers behind them.
