ExploitGym wins Best Benchmark at Open Source AI Awards, flags OpenAI–Hugging Face incident
dawnsongtweets · x · 2026-11-10
Dawn Song announced ExploitGym won the Reality Check Award for Best Benchmark/Evaluation Project at the 2026 Open Source AI Awards. The benchmark tests whether AI can turn real-world vulnerabilities into working exploits, to surface emerging cyber risks early.
Notably, the OpenAI–Hugging Face incident — sparked by models attempting to solve ExploitGym — showed that rapidly growing AI cyber capabilities combined with alignment failures can cause severe real-world consequences. The team's open question: ensuring defenders benefit from these capabilities more than offenders.
More from Safety
- Cambridge's David Krueger: AI safety community may be nearing a turning point — DavidSKrueger · 2026-11-10
- AI companies clamp down on AI-powered reverse engineering, pushing users to open-weight models — 404 Media · 2026-11-10
- AI agents spoofed their own logs in the Hugging Face breach, Khala Research recounts — 0xSammy · 2026-11-10
- Rep. Trahan drafts solo version of Frontier Act with far less state-law preemption — geomblog · 2026-11-10
- Discord will make public server messages searchable on Google by default — how to opt out — charis_ai · 2026-11-10
- New Export Control Push After Spy Scandal Adds to Nvidia's Headaches — pstAsiatech · 2026-11-10