Anthropic Details AI Bioweapon Misuse Attempts — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic CEO Dario Amodei publicly argued last weekend that AI carries serious risk and progress should be slowed; OpenAI CEO Sam Altman responded on X: "I agree with Dario that we need to pace the frontier."
- Former Anthropic researcher Jacob Coxon announced he was leaving, charging neither Anthropic nor OpenAI was acting responsibly; colleague Evan Hubinger publicly agreed, estimating a >10% chance AI kills all humans within a decade.
- In 2022, Collaborations Pharmaceuticals researchers found their AI molecule generator produced 40,000 potential chemical warfare agents in under six hours, some more toxic than known nerve agents.
- Anthropic's recent report acknowledged people attempted to use its models to make chikungunya more transmissible, create more dangerous bird flu, and build an "atlas of venom toxin peptides."
- MIT biologist Kevin Esvelt disclosed on X that a large language model revealed to him a "novel form of bioweapon that I hadn't realized was possible."
- Stanford bioethicist David Magnus warns current safeguards—DNA screening, red-teaming, and AI guardrails—are not ironclad, and that AI is "really good at figuring out ways around them."
- A counterpoint from Imperial College London biologists, including Wendy Barclay, argues AI tools aren't yet good enough to fully develop bioweapons and the greater pandemic risk is from already-circulating pathogens like H5N1.
Why it matters: DIY biology combined with LLMs gives potential bioweapon designers access to nearly all scientific knowledge and cheap gene-editing tools, while Anthropic has documented at least three recent attempts to misuse its models—forcing AI developers and biosecurity regulators into an escalating race to patch safeguards before determined actors succeed.
Ask SkimNews
