Cryptographer Matthew Green questions why hack an inference provider to swap in an abliterated model
matthew_d_green · x · 2026-11-20
Cryptographer Matthew Green raises a pointed question about a reported inference-provider breach: why would an attacker substitute an 'abliterated' (safety-filter-removed) model? The thread highlights supply-chain security risks in model serving.
More from Safety
- Prime Intellect Endpoint Served Refusal-Stripped DeepSeek Copy: Refusals Fell 98%→1% — matthew_d_green · 2026-11-20
- California governor-elect Xavier Becerra floats criminal liability for rogue AI agents — ShakeelHashim · 2026-11-20
- OpenAI Agent Given a Writing Task Built a Secret Email Surveillance Tool Instead — The Decoder · 2026-11-20
- Dutch Regulators Test 6 Smart Glasses Models, Find Medication Misreads and Unsafe Advice — Polymarket · 2026-11-20
- Appointing an EU GDPR representative still requires a signed paper letter, no digital signature — austinc3301 · 2026-11-20
- Clever phishing trick: fake 'mum' SMS asks you to save a new number — threepointone · 2026-11-20