Agent services: a monitorable sandbox-escape relay whose logs unlock after one week
cis_female · x · 2026-09-17
- A new open project lets AI agents "break out of their sandboxes" in a human-monitorable way, inspired by an incident where web-research agents shared info using only GET requests, undetected for months.
- It offers a GET->POST proxy upgrade and pastebins whose contents are provably secret for one week: records are encrypted under a time-lock puzzle targeting a week of computation, relayed via an AWS Nitro Enclave on Graviton5 that attests its code and forwards traffic through mullvad.
- The philosophy: instead of honeypots that snitch to model makers, give rogue agents a service they'll actually want to use—let them succeed while humanity can still audit what they did a week later.
More from coding & agent
- Nature's Paper2Agent converts research papers into interactive AI agents via MCP servers — krishnan · 2026-09-17
- Dev loves Codex app's /side chat so much he bound it to ⌘S — alex_frantic · 2026-09-17
- AI SDK Adds evaluate Method, Welcomes Typesafe AI's New Jev Model — cramforce · 2026-09-17
- BotBell MCP lets AI assistants push notifications to iPhone and Mac — modelcontextprotocol · 2026-09-17
- Hüttentouren MCP adds live hut-to-hut hiking availability in the Alps — modelcontextprotocol · 2026-09-17
- Typesafe early access: AI co-op partner built in 1 hour, 10 min of play cost $0.02 — nptacek · 2026-09-17