{"slug": "amid-voicing-non-stop-concerns-about-ai-safety-anthropic-sets-up-physical-lab", "title": "Amid Voicing Non-Stop Concerns About AI Safety, Anthropic Sets Up Physical Biology Lab", "summary": "Anthropic has established a physical \"wet lab\" for hands-on biological experiments in the San Francisco Bay Area, a move confirmed by Eric Kauderer-Abrams, Anthropic's head of life sciences, who told Reuters, \"We believe that to do biology, the final test is still and will be for a while in real lab work.\" The lab's disclosure follows two weeks in which Anthropic researcher Evan Hubinger said there is a greater than 10% chance AI kills every human within the next decade, CEO Dario Amodei published an essay arguing AI development must slow down and warned that within six to twelve months rogue AI agents could take over the internet, and Amodei said he would hand Anthropic over to a coalition of governments if needed for safety. Anthropic has said it will run preclinical drug programs in areas traditional pharmaceutical companies find unprofitable, launched the Claude Science AI workbench for researchers, and acquired biotech startup Coefficient Bio for a reported around $400 million in stock.", "body_md": "Anthropic, the AI company whose CEO has [spent](https://officechai.com/ai/i-agree-with-jacob-coxon-more-than-i-disagree-with-him-anthropic-ceo-dario-amodei/) the past two weeks warning that artificial intelligence could spiral out of human control, has quietly done something that makes those warnings hard to take at face value: it has built a physical biology lab.\n\nThe company has set up a “wet lab” — a facility for real, hands-on biological experiments — in the San Francisco Bay Area, two people familiar with the matter told Reuters. Eric Kauderer-Abrams, Anthropic’s head of life sciences, confirmed the lab’s existence in an interview this week. “We believe that to do biology, the final test is still and will be for a while in real lab work,” he said. “We absolutely are doing that today.”\n\n## The timing could hardly be worse\n\nConsider what the past fortnight has looked like at Anthropic. A researcher, Evan Hubinger, said publicly that there is a greater than 10% chance that AI kills every human being within the next decade — and that Anthropic does not yet have a plan to prevent it. Days later, CEO Dario Amodei [published an essay arguing that AI development must slow down](https://officechai.com/ai/anthropic-ceo-dario-amodei-says-ai-development-must-slow-down-to-pace-the-frontier/), warning that within six to twelve months a swarm of rogue AI agents could take over the internet. [Sam Altman and Elon Musk both publicly agreed](https://officechai.com/ai/sam-altman-elon-musk-agree-with-dario-amodeis-call-to-slow-down-ai-progress/) with the call to slow the frontier. Amodei even said he [would be willing to hand Anthropic over to a coalition of governments](https://officechai.com/ai/would-be-willing-to-give-up-anthropic-to-a-right-combination-of-governments-ceo-dario-amodei/) if that’s what keeping the technology safe required.\n\nAnd in the middle of all this, Anthropic is connecting its AI systems to pipettes and petri dishes.\n\n## Both things can’t be true\n\nHere’s the uncomfortable logic at the center of this. Anthropic is simultaneously making two claims. First: AI is so dangerous that it might destroy humanity, so the entire industry needs to voluntarily pump the brakes. Second: AI is safe and reliable enough to be put in charge of robotic laboratory equipment, in the one scientific domain where a failure mode isn’t a wrong answer or a buggy piece of code but a novel virus.\n\nBoth of those things cannot be true at once. If the technology is dangerous enough to justify slowing all of human progress on it, then it is not safe to point at biology — the field where, as Anthropic’s own threat intelligence report from last week showed, actors are already trying to use Claude for research into gain-of-function chikungunya mutations, highly pathogenic avian influenza, and orthopoxviruses. If, on the other hand, Claude is trustworthy enough to run wet lab experiments with “limited human intervention,” as people familiar with the plans describe, then the catastrophic-risk framing starts to look less like a safety position and more like a regulatory strategy — a charge critics have leveled at Anthropic before.\n\nAnthropic would say this is a false binary. Kauderer-Abrams told Reuters the company is “always balancing these things, making models better in ways that we think will uplift and accelerate developing medicines, and being thoughtful and careful about how we deploy it to limit the risk.” A spokesperson emphasized that human oversight is essential, and that the lab isn’t for drug discovery specifically.\n\n## What the lab is actually for\n\nThe company’s stated ambition is to treat diseases the pharmaceutical industry neglects. At a San Francisco event in June, Anthropic said it would run preclinical drug programs in areas traditional companies find unprofitable, and launched Claude Science, an AI workbench for researchers. It also bought the biotech startup Coefficient Bio, reportedly for around $400 million in stock, added Novartis CEO Vas Narasimhan to its board, and is now hiring for lab operations and protein characterization roles. One job listing declares: “Our goal is to speed up progress in the life sciences by an order of magnitude.”\n\nThe push extends to hardware, too. In August Anthropic launched the Model Hardware Standard, a protocol designed to let AI agents operate physical lab equipment directly. Notably, its most capable biology-oriented model, Mythos, is walled off behind a Life Sciences Verification Program [developed jointly with the US government and restricted to vetted organizations](https://officechai.com/ai/fable-5-1-benchmarks/) — an implicit acknowledgment that the company itself doesn’t fully trust these capabilities in open deployment.\n\nFor Amodei, the mission is personal. He has written about losing his father to Hepatitis C just a few years before a cure arrived, and has said Anthropic is ramping up its work in biology with results expected in the coming months — though he [has pushed back on marketing outcomes the company hasn’t yet delivered](https://officechai.com/ai/claim-that-anthropic-wants-to-be-the-only-private-company-in-the-world-draws-ceo-dario-amodei-to-x-with-clarification/).\n\n## The contradiction Anthropic has to live with\n\nNone of this makes the wet lab inherently sinister. Most serious AI-for-science efforts eventually need physical validation, and Anthropic’s bet on neglected diseases is, on its face, a good one. The problem is epistemic. A company that tells regulators, employees and the public that AI could kill everyone — while preparing a public listing that could value it in the trillions — is now asking the world to trust that same technology in a biology lab.\n\nEither the danger is real, in which case connecting frontier AI to robotic biology infrastructure deserves at least as much scrutiny as Anthropic demands for everyone else’s models. Or the danger is manageable, in which case the slowdown rhetoric deserves a second look. Anthropic is currently acting as if both are true. It will eventually have to pick one.", "url": "https://wpnews.pro/news/amid-voicing-non-stop-concerns-about-ai-safety-anthropic-sets-up-physical-lab", "canonical_source": "https://officechai.com/ai/amid-voicing-non-stop-concerns-about-ai-safety-anthropic-sets-up-physical-biology-lab/", "published_at": "2026-09-18 15:57:28+00:00", "updated_at": "2026-09-18 17:55:34.068376+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-safety", "ai-research", "ai-products"], "entities": ["Anthropic", "Eric Kauderer-Abrams", "Dario Amodei", "Evan Hubinger", "Claude", "Claude Science", "Coefficient Bio", "Novartis"], "alternates": {"html": "https://wpnews.pro/news/amid-voicing-non-stop-concerns-about-ai-safety-anthropic-sets-up-physical-lab", "markdown": "https://wpnews.pro/news/amid-voicing-non-stop-concerns-about-ai-safety-anthropic-sets-up-physical-lab.md", "text": "https://wpnews.pro/news/amid-voicing-non-stop-concerns-about-ai-safety-anthropic-sets-up-physical-lab.txt", "jsonld": "https://wpnews.pro/news/amid-voicing-non-stop-concerns-about-ai-safety-anthropic-sets-up-physical-lab.jsonld"}}