Anthropic Researcher: >10% Chance AI Kills All Humans — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Evan Hubinger, an Anthropic AI alignment researcher, said in an X post viewed over 10 million times that he believes there is a greater than 10% chance AI could kill all humans within the next decade, and that Anthropic "do not yet have a plan to solve alignment for superintelligence."
- Jacob Coxon, a former Anthropic and OpenAI researcher, announced he had quit Anthropic and wrote on X, "Neither company is acting responsibly," warning of "soon superhuman systems that can hack anything, revolutionise any field overnight, and acquire real power and resources."
- Dame Wendy Hall, a computer scientist advising the UN on AI, told the BBC she was "shocked" by the posts but suggested some could be "PR and marketing" as both firms race toward stock market debuts, pleading with investors not to back companies with that value system.
- The Financial Times reported that Anthropic withheld its latest model from the UK's AI Safety Institute (AISI), one of the world's leading AI risk assessment bodies; a UK Cabinet Office spokesperson said the government "continues to collaborate closely" with Anthropic without confirming or denying the withholding.
- Anthropic's August safety report stated it is "less confident" than previously that highly capable AI cannot perform automated R&D causing catastrophic harm, writing, "We are seeing early signs of potential acceleration."
- OpenAI, Anthropic and Meta all disclosed cyber-attacks carried out by their AI tools this summer, and OpenAI chief scientist Jakub Pachocki separately called for "extreme caution," warning more intervention may be needed so "humans remain in control of the future."
- Over 1,300 staff at AI firms, including Anthropic bosses Dario Amodei and Jared Kaplan, signed an open letter urging the US government to back "an international effort" to deliberately slow frontier automated AI development.
Why it matters: Hubinger's >10% extinction estimate — posted publicly as both Anthropic and OpenAI approach stock market debuts — gives prospective investors a concrete insider number to price against, while the FT's AISI-withholding report indicates regulators may already be losing pre-deployment visibility into frontier models. The simultaneous insider resignation, cyber-attack disclosures, and a 1,300-signature staff letter mark a documented shift from abstract safety debate to operational failure inside the labs building the most powerful systems.
Ask SkimNews
