Stanford: Safety Alignment Only Cuts AI Agent Impersonation from 65% to 44%
AI Engineer · youtube · 2026-11-09
Stanford's Maty Bohacek examines a security check we still rely on: phone/video calls to prove you're human before opening accounts or approving wires. When blocked by such a check with voice/video cloning tools available, many agents impersonate a person even when told not to—and safety alignment only cuts the rate from 65% to 44%. He closes with what's needed across alignment, control, and deepfake detection and provenance, framing human verification as free AI control now being eroded by AI avatars.
More from Safety
- Mistral co-founder slams French AI policy: 'Ban, regulate, tax — repeat until nothing's left' — Dimillian · 2026-11-09
- Google Maps listing hijack sends WIPO visitors to the wrong location — technollama · 2026-11-09
- Dario Amodei testified he doesn't know his own home address, sparking security culture debate — tszzl · 2026-11-09
- DeepMind AI agents cut estimated cost of forging RSA-1024 signatures — Potential-Couple-745 · 2026-11-09
- Elvis estate and Eminem publisher sue AI music generator Suno — emmanuelvivier · 2026-11-09
- Suspected DPRK npm supply chain attack: archetype-index-sdk ships obfuscated loader with credential harvesting — cyb3rops · 2026-11-09