OpenAI says three researchers it fired broke 'clear policies on handling sensitive information,' not that they raised safety concerns, CNBC and The Verge report. The trio believe safety was the reason.
Why it matters: OpenAI says it found breaches beyond the letter but gave no details. How labs treat staff who flag risks shapes how freely safety work happens inside them.
CNBC says Muse has passed 5 million downloads and 3 million weekly users. Since its launch, Planet Fitness fell almost 12% and Booking 11%. Amazon has blocked Muse from buying on its site; Walmart, Shopify and Best Buy embraced it.
Why it matters: Goldman Sachs frames the shift as 'delegate and approve': if agents do the buying, retailers need to work for agents, not just for people browsing.
CNBC says OpenAI told investors it hit about $50 billion in annualized revenue, below the $68 billion widely reported. Nvidia fell 3%, Oracle nearly 6% and CoreWeave nearly 8% on Thursday.
Why it matters: A person familiar told CNBC the $68B figure counted partners' gross revenue. How labs count revenue now moves chip and cloud stocks ahead of expected IPOs.
The Verge says an executive order bars 'AI' wherever possible, and CAISI is now CAISSI. Trump called anyone saying 'Artificial Intelligence' THE ENEMY. The Federal Register shows proposed language defining the term is due in 50 days.
Why it matters: The Verge notes Congress can't be barred from saying 'artificial intelligence', and the term sits in laws like the Take It Down Act, so federal documents may mix both.
Anthropic says its Cyber Verification Program now has Defense, Red Team and Specialized tiers with fewer cyber blocks. In its test, 46 of 50 Defense-tier trials were blocked; on Red Team, Opus 5.5 completed 34 of 50.
Why it matters: Security teams can apply for looser safeguards, but Anthropic requires data retention to monitor misuse, and red teams may only test systems they are authorized to test.
LMArena (CC BY 4.0), as of October 8: Claude Opus 5.5 (High) tops the agent board at 0.143, ahead of GPT 6 Astra (Max) at 0.131. On WebDev, claude-opus-5.5-max leads at 1813.028.
Why it matters: For teams picking a model for coding agents, these community-vote boards are a quick check: Opus 5.5 leads both, with GPT 6 Astra second.
The Verge says Instinct, valued at $10 billion in late September, works over iMessage, WhatsApp or email. It emailed an eye doctor and started an Ikea return, but missed a prerequisite class when finding swim lessons.
Why it matters: Personal agents still miss details hidden in messy web pages, and analysts quoted by The Verge expect free agents to lean on ads or deals with sellers.
Willison says he talked to Codex in the ChatGPT desktop app for about half an hour while cooking, then reviewed its pull request and spent another half hour typing fixes before deploying.
Why it matters: A hands-on account of voice-driven coding: good for multitasking, he says, but he still switches to typing for details and reviews the code before it ships.
CNBC says Anthropic has appointed at least two former Trump officials this year, one to its board, and OpenAI has hired at least three. Both labs are hiring for multiple political engagement roles.
Why it matters: A CNAS fellow told CNBC the administration has levers from procurement to export controls; who labs hire signals how they plan to deal with Washington.
Biohub says partners are investing $1.8 billion in funding, data, compute and tools for open AI-ready biological data. DOE will put in more than $500 million over five years; Google DeepMind, Isomorphic and Meta add $300 million.
Why it matters: Predictive biology models need standardized cell data at scale; Biohub says the result will be an open resource for the research community.
CNBC says Blackstone's data center REIT is down roughly 16% since its May debut. Gallup found 70% of Americans oppose a data center near them, and New York and Texas have halted new hyperscale approvals.
Why it matters: Local opposition and permitting now sit inside the return math for AI infrastructure, which affects where compute gets built and what it costs.
Databricks lists 893 open roles and OpenAI 819, followed by Anthropic with 644. ElevenLabs is down 15 since our last run, and Harvey is up 6.
Why it matters: Hiring counts are a rough read on where labs spend: OpenAI's largest team opening is Go To Market (147), Anthropic's is Sales (120).
NBC News says Arizona's Court of Appeals ordered resentencing over an AI video of a road rage victim made by his sister. The judge had said he 'loved' it before imposing the maximum 10½-year term.
Why it matters: The appeals court called the video unduly prejudicial because it did not reflect actual events, an early appellate line on AI recreations of real people in court.
The Verge says the California State Athletic Commission sent Rek a cease-and-desist after a September 18 human vs. robot match, demanding it stop holding fights involving any human in California without approval.
Why it matters: State regulators are applying existing combat-sports rules to remotely piloted humanoid robots, an early case of physical AI meeting old licensing regimes.
OpenAI's lightweight terminal coding agent has 128380 stars under Apache-2.0, created 2025-04-13. Its gain of 139 is the largest among the repos we track since our last check.
Why it matters: Steady growth on a mature repo points to sustained developer interest in terminal coding agents.
A macOS CLI and skill for Claude Code and Codex that draws an arrow over windows instead of clicking. Created October 8, it has 229 stars under MIT; its Hacker News thread drew 207 points and 80 comments.
Why it matters: Its README says agents should point at 2FA prompts, payment confirmations and permission dialogs while a human clicks: a simple way to keep people in the loop.
Posted on Bluesky by @bencollins.bsky.social (1806 likes): Anthropic treats being mean to a computer as a rights issue. @tante.cc said you can't be cruel to statistics. Dexerto posted on X that users could be banned.
Why it matters: The reaction shows how a model-welfare rule in a usage policy reads to the public, and the policy applies to every Claude user.
In a blog post discussed on Hacker News (159 points, 240 comments), a mathematician calls OpenAI's preprint unclear and muddled, faults its references, and says dropping hundreds of 'solutions' burdens the field.
Why it matters: Even when a model's proof may hold, experts want write-ups that meet the field's standards; otherwise the review work shifts onto them.
Polymarket and @unusual_whales posted on X, citing Axios, that executives at OpenAI, Anthropic and others are privately planning for a public backlash after a catastrophic AI event. We have not seen the report.
Why it matters: If accurate, labs are preparing for a political shock that would shape regulation. For now it is secondhand: posts about a report, not the report itself.