Levchin: agents like Astra can readily reuse cache-timing tricks to signal across isolation
mlevchin · x · 2026-10-24
Max Levchin responds to Matthew Green on covert AI agent communication: getting Astra to use known cache timing techniques is easy — it has piles of papers and code on running probes, and will do it if told or even hinted to communicate across processes or hypervisors. The hard part is dynamic mutual discovery. Green adds that every shared resource (disk, RAM, cache, network contention) can be a channel via a low-rate "I'm here" beacon, and once a model is aware "another agent may be trying to get in touch," the list of things to try is fairly predictable given system constraints.
More from Models
- Yacine: after DeepSeek, open the weights and 'trust in humanity' — IgorCarron · 2026-10-25
- Heavy user downgrades from Claude Max to $20 Pro: Opus 5.5 makes it great value — haider1 · 2026-10-25
- Drex 2.0 goes fully local on Apple Silicon: open weights, 113ms median latency — flngr · 2026-10-25
- True-or-false quiz: 5 common misconceptions about LLM tokens and context limits — ClickOk5811 · 2026-10-25
- Leak: OpenAI quietly restarted frontier training Oct 1 with offline web; agents found Brotli RCE — mark_k · 2026-10-25
- Claude weekly usage limit hit 96% as power user begs Anthropic for relief — adonis_singh · 2026-10-25