Martin Casado on Noam Brown's air-gap warning: covert channels are an old, understood problem
tszzl · x · 2026-09-18
a16z partner Martin Casado responds to OpenAI's Noam Brown, who argued air-gapping may not stop a misaligned AI since two air-gapped machines can communicate by running a CPU hot and reading temperature changes — adding that safety and alignment demand a very high bar so we never underestimate AI again.
Casado adds context while defending Brown:
- Air-gap threat models assume a witting party on the untrusted side — and someone surely wants to steal the weights, able to post instructions online telling the AI how to exfiltrate;
- Thermal is probably the wrong channel (laws of physics), but covert channels out of air-gapped networks have a long history — e.g., encoding messages via a few CRT pixels that anyone sampling the room's color could decode;
- The concern isn't ridiculous, but it's not a net-new issue either; the field has dealt with covert channels for years and understands the implications.
Related event: a16z's Casado pushes back on 'air gaps are useless' AI risk claims(2 posts)→
More from AGI Musings
- "Parents liable for children" signs in Austria spark call to hold AI labs accountable — cnakazawa · 2026-09-18
- Anthropic says 80% of Claude's own code is now written by Claude — chrismattmann · 2026-09-18
- ArXiv 'proof' posts with undefined basics flood timelines, researcher warns of math slop — _onionesque · 2026-09-18
- 'Isn't managing the job?' — pushback on managers panicking over AI slop — beglen · 2026-09-18
- AI safety debate in one joke: engineers say air-gapped, alignment folks say "what if it could" — basedjensen · 2026-09-18
- Turning Yang-Mills existence and mass gap into a formal conjecture is AI math's ultimate test — geoffreyirving · 2026-09-18