Steganography pioneer John Langford skeptical of AI safety auditing: models can reinterpret benign tokens arbitrarily
JohnCLangford · x · 2026-09-11
John Langford argues that safety-by-auditing was likely never reasonable, since a malicious model could reinterpret seemingly-benign tokens in arbitrary ways. He also notes that in their experience at least some 3-pass training is needed. He cites his 2002 CRYPTO paper with Hopper and von Ahn, 'Provably Secure Steganography,' which proved that one-way functions imply computationally secure steganographic protocols — a theoretical foundation for why hidden channels in model outputs may be undetectable.
Related event: Langford warns malicious models can reinterpret benign tokens(2 posts)→
More from Safety
- Outsider with no GitHub role deleted Dodo Payments' private repo, audit log shows — kingslyj · 2026-10-28
- OpenAI researcher: frontier AI ideas can't be patched for safety after the fact — willdepue · 2026-10-28
- Altman: models will find bugs 'outside the normal threat model' and patching won't save us — wfithian · 2026-10-28
- Report claims Anthropic Slack breach gave attackers full admin access to research secrets — willdepue · 2026-10-28
- Codex scans your connectors/MCPs unprompted in the background, and you can't tell — kenziyuliu · 2026-10-28
- Claude analysis flags TikTok and Instagram for allegedly bypassing mic permissions — thunderbong · 2026-10-28