Anthropic is operating a Bay Area wet biology lab that uses its AI models to run physical experiments, confirmed by TechCrunch. The lab has been running “today,” and Anthropic also bought Coefficient Bio in April to support its AI biotech work. The company is positioning the lab around fundamental biology while expanding access through its Life Sciences Verification Program and continuing partnerships rather than direct drug discovery.
Anthropic said Accenture staff will work inside its lab to scrutinize the company’s models and personnel as part of embedded evaluation. Accenture’s Faculty unit will begin evaluating and red-teaming models, and the two companies expect to invest at least $1 billion over the next five years. This adds a new third-party embedded evaluator and expands external model-safety testing beyond the usual outside reviews, with Anthropic noting its approach will evolve as no standards exist yet.
Anthropic is partnering with Accenture to run independent embedded evaluation of frontier AI, including red-teaming and alignment and safeguards testing. The companies expect to invest at least $1 billion over the next five years to build evaluation capacity. This will add evaluators with near-employee access inside labs, producing incident reporting and a more verifiable public account of model safety while Anthropic keeps accountability.
Anthropic PBC opened a wet lab in the San Francisco Bay Area to use robots for biology research tasks. The robots will be powered by Claude, and Anthropic reported Claude Mythos 5.1 generated 12 candidate molecules with a 50% hit rate. The lab will likely use Claude with its MHS protocol to automate lab-instrument workflows and may involve external partners for some work.
Simon Willison’s Weblog·3 weeks ago·
40
● 2 sources
Claude Code added support for AGENTS.md so it can fall back to that file when CLAUDE.md is missing in a folder. It starts in version 2.1.277. This changes how project instructions are discovered and lets users customize instructions via built-in and future mods.
Hacktron AI security researchers used Claude Opus 4.8 to turn a libheif memory-corruption bug into a working ARM64 exploit against a Mac and then achieved remote code execution on a test forum, before escalating to proof of access in OpenAI’s private monorepo via an employee Codex account. Roughly 72 hours after they began, they read OpenAI’s private openai/openai monorepo and opened a README pull request. OpenAI narrowed permissions for community sign-in tokens and revoked the affected tokens and sessions, while the incident also prompted revisions based on the new Opus 5 capability in exploit development.
Dario Amodei outlined a plan for Anthropic to “pace the frontier” using independent safety evaluators and coordination between AI labs in democratic countries. A week after an Anthropic researcher’s doomsday warning, the proposal drew industry support and Jensen Huang’s pushback. The discussion shifts to whether labs can agree on what “slowing down” means and who would enforce it.
TechCrunch’s Equity podcast discusses Dario Amodei’s plan to “pace the frontier” of AI development using independent safety evaluators and coordination among AI labs. The episode highlights Automattic’s boardroom turmoil tied to Matt Mullenweg’s 33-hour ouster. It shifts the focus to whether labs can agree on what “slowing down” means and who should enforce it, while also recapping major tech deals mentioned in the show.
Hacktron AI researchers used Anthropic’s Claude to chain two vulnerabilities and gain access to multiple OpenAI employee ChatGPT accounts through OpenAI’s Discourse forum. They received a $6,500 bug-bounty award after reporting the issue, and OpenAI says it has resolved the problems. The incident shows model-assisted hacking working with off-the-shelf tools and highlights how fixes that lack formal vulnerability tracking can linger in third-party software.
Open-weight models took a majority of tokens routed through Vercel’s AI Gateway, reaching 56% of monthly token volume for the first time. Anthropic still accounted for 64% of the estimated dollars spent via the gateway in August, even though open-weight models processed 56% of the tokens. As open-weight usage grows and average per-token pricing declines (down 23.2% in August), spend remains concentrated with Anthropic while teams shift within and across model labs based on cost and consistency.
Steve Eisman said leading AI companies are trying to create a crisis to benefit themselves, dismissing AI doomsday warnings as self-serving marketing. He pointed to an IPO that Anthropic has already “confidentially filed” for. The debate over AI safety slowdowns shifts toward viewing regulation and competitive positioning, rather than an imminent AI catastrophe.
AINews reported that its weekend scan found no additional Discord updates and otherwise compiled a roundup of AI agent productization, tooling, research, benchmarking, and security discussions across posts and releases. The standout quantitative detail was Anthropic’s claim that the Claude-led share of model R&D tasks rose from 1% to 26% in about 6 months. The coverage shifts from scattered “chat + tools” concepts toward orchestrated multi-session agents with managed harnesses, clearer eval methods, and more concrete internal measurement and security lessons.
Hard Fork ended after focusing for most of its run on understanding how LLMs and AI safety affect companies and everyday life. The final show wrapped up ahead of Machine Gods launching the week of October 19, with episodes planned twice a week. The shift moves coverage from the New York Times podcast to a new Kevin Roose/NPR project under Machine Gods Media, including expanded YouTube output and more frequent publishing.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
