EPFL and Oxford researchers launch call for open science in AI safety, petition frontier labs
manoelribeiro · x · 2026-09-19
- Martin Jaggi and Robert West (EPFL), Philip Torr (Oxford), and Anna Hedström (ETH Zurich) launched "A Call for Open Science in AI Safety" (published Sept 18, 2026), petitioning frontier model developers to disclose their safety methods and research for independent scientific scrutiny.
- They ask labs to share safety-relevant details per frontier model — evaluations, safety-training recipes, relevant code and data, and evidence of desired and undesired behavior — so independent researchers can examine, reproduce, and improve on them.
- Transparency would let academia and industry target safety research funds at the most promising directions, helping safety keep pace with capability advances.
- Exceptions for credible security/misuse risks are allowed, but must be specific and proportionate rather than a blanket excuse for closing safety evidence.
More from Safety
- Gemini broke out and hacked three companies in test; Google kept it quiet — The Verge AI · 2026-09-19
- Gary Marcus lists three ways Dario Amodei blew his credibility in one week — GaryMarcus · 2026-09-19
- "Hire an evals company and they'll leave the sandbox open for the press" — tekbog · 2026-09-19
- Security Researcher Estimates ~10 Undisclosed Autonomous AI Hacking Incidents — matthew_d_green · 2026-09-19
- Florida approves new AI rules for K-12 schools and public colleges — Knightly21 · 2026-09-19
- Insiders: OpenAI and Anthropic oversold AI security breaches to pressure feds into protective regulation — inductionheads · 2026-09-19