Matthew Green hints abliterated GLM found an exploitable pre-auth bug in 2 hours
MalwareTechBlog · x · 2026-11-09
Cryptographer Matthew Green hinted in a spat with Marcus Hutchins (MalwareTech) that weeks ago he used an "abliterated" (guardrail-stripped) GLM model to find a pre-auth remotely exploitable vulnerability in just 2 hours, which the vendor confirmed and moved into coordinated disclosure.
Hutchins fired back that it's "just words," demanding Green actually drop a zero-day to prove claims that low-skilled threat actors can effortlessly mine vulnerabilities with LLMs — noting Green has 0 CVEs. The dispute centers on whether uncensored open-weight models meaningfully lower the bar for vuln discovery.
More from Fun
- If you're not donating, it's not effective altruism, it's just sparkling philosophy — davidmanheim · 2026-11-09
- 'Is Aella neurotic?' locked after 12,239 pages — forum history's greatest thread — mimi10v3 · 2026-11-09
- China's comfy brain-reading device hits 94% typing accuracy in day-long trial — AIFlow_ML · 2026-11-09
- Jensen Huang's NVDA stake worth ~$151.5B — user does the math on a hypothetical divorce selloff — 0xSammy · 2026-11-09
- Dev builds an oddly addictive volume slider UI people can't stop playing with — sebkrier · 2026-11-09
- Bassist builds agent-readable audio engine catalog to stop vibe-coded plugin sameness — jameshgordy · 2026-11-09