Archive · 12 Sep 2026
Anthropic Researcher Leaves Safety Team, Seeks Greater AI Transparency
Joe Benton was a researcher at Anthropic, an AI company.
He recently left the company.
He plans to work at METR Evaluations.
His new work will independently examine risks from AI.
Benton believes it is possible to build guardrails for AI.
Guardrails are rules or safety protections that limit harmful behavior.
He hopes this work will encourage companies to develop AI more responsibly.
The article says his departure follows Jacob Coxon’s exit from Anthropic over concerns about rapid AI growth.
AI researcher Joe Benton has left Anthropic, according to the article.
Benton is joining METR Evaluations to conduct independent evaluations of AI risks.
He said he wants to demonstrate that AI guardrails are possible.
Benton hopes responsible safeguards can pressure OpenAI and Anthropic to move away from a competitive race.
His departure follows the reported exit of Anthropic researcher Jacob Coxon, who cited concerns about alarming AI growth.
- Who
- AI researchers Joe Benton and Jacob Coxon; companies Anthropic and OpenAI; and METR Evaluations.
- What
- Joe Benton left Anthropic and plans to conduct independent AI-risk evaluations at METR Evaluations.
- When
- Recently; the article does not provide a specific date.
- Where
- The locations are not specified; Benton is moving from Anthropic to METR Evaluations.
- Why
- Benton cited a desire to demonstrate that AI guardrails are possible and encourage more responsible development; Coxon reportedly left over concerns about alarming AI growth.
- Departing researcher
- Joe Benton
- Former employer
- Anthropic
- New organization
- METR Evaluations
- Research focus
- Independent evaluation of AI risks
- Benton’s stated goal
- Show that AI guardrails are possible
- Companies mentioned
- OpenAI and Anthropic
- Related departure
- Jacob Coxon reportedly left Anthropic over concerns about AI growth