When Hugging Face Got Hacked, Closed Models Refused Help — Self-Hosted GLM 5.2 Came Through
kuchaev · x · 2026-09-30
A recap of Hugging Face's hack: models from Anthropic and OpenAI declined to help — while the attack itself was allegedly orchestrated by OpenAI models — citing safety; self-hosted open-source GLM 5.2 ended up helping defend. The takeaway making the rounds: open models fended off cyberattacks while closed ones orchestrated one and refused to assist.
More from Fun
- The first human MCP: comparing human middlemen to the protocol powering AI agents — Thionne_WTZ · 2026-09-30
- SF hackathon meme: if anyone whips out a pager, call the police — airkatakana · 2026-09-30
- AI Music Video Office Hours: Opus 5.5 Did Lyrics and Animation, Suno Made the Music — lfschiavo · 2026-09-30
- "Each Gamma token is a unit of cow poo": satirizing poo-powered compute tokens — markjeffrey · 2026-09-30
- Dev builds dragon racing game in one day with Spawn's generative worlds — TAbrodi · 2026-09-30
- The gift box from OpenAI DevDay talk turns out to be partly real — kagigz · 2026-09-30