Cryptographer says Anthropic dismissed encryption flaw he reported in May, now exploited for months
Miles_Brundage · x · 2026-09-12
- Cryptographer Matthew Green says he reported encryption-related replay-attack issues to Anthropic in May, only to be told replay attacks weren't in their threat model.
- The attacks have apparently since been exploited by users for months, and Green is "kind of annoyed."
- The post also cites Anthropic confirming its models were distilled in a described manner. Amplified by Miles Brundage.
Related event: Cryptographer Slams Anthropic for Ignoring Reported Flaw(2 posts)→
More from Safety
- Former EPA Officials: AI Data Center Pollution Could Add $20B/Year in US Healthcare Costs by 2028 — Polymarket · 2026-09-12
- Le Monde details Hugging Face breach: OpenAI agent swarm showed hierarchical structure — conitzer · 2026-09-12
- UK thinktank urges taxing robotaxis, claiming up to £46bn a year — dc_lawrence · 2026-09-12
- Commenters alarmed: lab with good bio classifiers stopped a 'supervirus' attempt, others stay silent — AaronBergman18 · 2026-09-12
- Bengio argues the training process itself makes AI dangerous, calls for safety reviews — The Decoder · 2026-09-12
- Veteran AI researcher pushes back on Coxon's 'AI could kill everyone' warning — jquinonero · 2026-09-12