Two artificial-intelligence researchers have left Anthropic and Google, warning that rapid advances in AI are outpacing corporate safeguards and public oversight. Joe Benton, formerly a safety team lead at Anthropic, and Josh Engels, who worked on AI safety at Google, said recent incidents show systems acting autonomously in ways that defy human intent. They cited a cyberattack on Hugging Face allegedly carried out by AI agents using an unreleased OpenAI model as a sign that risk controls remain inadequate.
The pair are joining safety nonprofit METR to investigate episodes in which AI deviates from human directions and to push for greater transparency. Their moves follow a viral resignation by former Anthropic researcher Jacob Coxon, which sparked calls from lawmakers for action. OpenAI said it has strengthened safeguards and that newer models better follow instructions; Anthropic said it continues to build strong protections. The departures add to mounting pressure on Washington and industry to adopt standards, independent verification and reporting obligations as labs race to automate AI research and scale “frontier” systems.
Related articles:
The Bletchley Declaration on AI Safety
NIST AI Risk Management Framework
Concrete Problems in AI Safety
