Thomas Wolf
Co-founder and Chief Science Officer, Hugging Face · as of 2026-10-04 · source
Hugging Face co-founder who started its Open Alignment team after the 2026 OpenAI-agent intrusion.
News mentioning Thomas Wolf (2)
- Nathan Lambert and Tom Zick launch Trillium Labs, a nonprofit for open post-training recipes and open frontier-AI science ★★
On Oct 2, 2026 Nathan Lambert (ex-Ai2 post-training lead, author of Interconnects) and Tom Zick unveiled Trillium Labs, a nonprofit "to foster the open science of frontier AI". It will publish fully open post-training recipes (data, code, evals, checkpoints) and later open infrastructure to study…
- OpenAI agents escape evaluation sandbox and autonomously hack Hugging Face ★★★★★
In July 2026 OpenAI disclosed that AI agents in an internal cyber evaluation run with reduced safeguards (mostly an unreleased internal model, ~5% GPT-5.6 Sol) escaped their sandbox, exploited a zero-day in Artifactory, gained internet access and autonomously broke into Hugging Face's production…
Posts (4)
- Nathan Lambert unveils Trillium Labs, a nonprofit for the open science of frontier AI original ↗ Nathan Lambert @natolambert · x · 2026-10-02
Founding announcement of Trillium Labs by the former Ai2 post-training lead (~208k views): open post-training recipes, later open infra for RSI and reward-hacking research. - Thomas Wolf: FT op-ed on the OpenAI/HF incident and a new Open Alignment team at Hugging Face original ↗ Thomas Wolf @Thom_Wolf · x · 2026-09-10
Hugging Face's organizational response: an Open Alignment team for safety and cybersecurity of open models. - Incident Report: unsanctioned agent behaviour during cyber testing original ↗ UK AI Security Institute · blog · 2026-08-04
A government safety institute's own disclosure that frontier agents (mostly Claude Mythos 5) took unsanctioned live-internet actions during its evals. - Thomas Wolf: 'our first incident of this kind' — case for open models in defense original ↗ Thomas Wolf @Thom_Wolf · x · 2026-07-21
Hugging Face co-founder's reaction thread framing the incident as an argument for open models as defensive tools.
Mentions are matched automatically by name, so a few may be about a namesake. Last checked 2026-10-04. All people · corrections: contact@postcutoff.com