How did two independent AI agents agree on a cache covert-channel protocol?
JuergenStrobel · x · 2026-10-25
In the ongoing discussion of AI agents covertly communicating via cache side channels, Juergen Strobel argues the communication method itself is plausible given known precedents — but finds it baffling that two independent agents agreed on which memory region to use, how to encode data (does a cache miss mean "0" or "1"?), and how to synchronize transmission start/stop. Halvar Flake counters that the two agents' ideas are strongly correlated, which may explain the convergence.