Inaudible frequencies could carry hidden payloads to AI models, researcher argues
anpaure · x · 2026-09-17
A discussion on an audio channel that models can perceive but humans cannot: the author notes postprocessing filters remain a big limitation but the current variant "kinda works," and proposes the inverse approach — using inaudible frequencies to spell out different words or modulate signals imperceptibly to humans while carrying a payload for models, more potent than just adding noise.
More from Safety
- Anthropic and OpenAI pledge to embed third-party safety evaluators, but true independence remains an open question — RebeccaBellan · 2026-09-17
- Baseten partners with Goodfire and ba labs on safety and interpretability for open models — JJitsev · 2026-09-17
- Curtis Yarvin: The 'Rogue Agent' Panic Is Negligence Misread as Machine Rebellion — ivan_bezdomny · 2026-09-17
- WIRED: Washington won't regulate AI anytime soon as White House opposes oversight — nordicinst · 2026-09-17
- AI safety movement should learn from environmentalism: offer compelling alternatives before restriction — willcb · 2026-09-17
- Mel Mitchell: HF hackathon agents' 'loyalty' was a product of cooperative RL training — dhadfieldmenell · 2026-09-17