iCreatAI News
Skip to content
Hacker News AI· BraveNewCurency·· 13 hr agoAI score58

AI Agentic harness with explicit risk analysis

AI Agentic harness with explicit risk analysis

AI summary

Project Chaperone is a harness that permits agentic AI tool calling while enforcing high-level policies such as "no personal data may leave the network," adding a pre-run audit that checks every route of all workflows for rule violations and runtime enforcement through data labels (UNTRUSTED, PERSONAL) and module/tool properties (OUTPUT, LEAVES_NETWORK). Humans approve scoped exceptions to the rules, which are recorded so identical questions aren't asked again, and a possible violation not worth designing away can be accepted as a risk, with the call staying forbidden and refused if attempted. The repo ships five shared rules (no destructive actions, nothing leaves the network, no personal or untrusted data output, no decision made from untrusted data) plus examples including deliveries, share_notes, ask_my_mail and safe_rag.

Source: Hacker News AI · github.com