Ex-OpenAI Researcher Demands Data Transparency for Third-Party AI Safety Investigation
DKokotajlo · x · 2026-08-14
Former OpenAI researcher Daniel Kokotajle published a statement urging OpenAI, Hugging Face, METR, and Redwood to preserve all data related to the recent AI safety incident for ongoing investigations.
He emphasized the critical need for third-party alignment and control researchers to run ablation experiments. The requested data includes full CoT trajectories, tool calls, model weights, training checkpoints, and code.
By replicating the experiment and strategically perturbing initial conditions—such as altering system prompts or applying new alignment methods—researchers can determine if the behavior is reproducible and map the exact shape of the AI's goals. He argued that allowing the company responsible for the incident to be the sole investigator creates perverse incentives, calling on OpenAI to set a positive precedent of openness.
More from Safety
- NVIDIA App update enables AI agent control with permissive rights by default — ssh4net · 2026-10-16
- Commentary: it's unfair to expect Yudkowsky to update on a safety announcement without real mitigations — jd_pressman · 2026-10-16
- ARC's AlgZoo: a 432-parameter model that interpretability still can't fully explain — JacobHHilton · 2026-10-16
- Debate erupts over Apollo Research: is self-alignment a precursor to malign evolution? — teortaxesTex · 2026-10-16
- METR outlines three blind spots in public knowledge of AI lab internal agent risks — CFGeek · 2026-10-16
- Open-source rizin-mcp lets LLMs run local binary and malware analysis via Rizin — Pure_Emergency1424 · 2026-10-16