Ask HN: AI agent acts. Your logs say so. But you wrote the logs

5 points by Yahyaaa 36 days ago | 6 comments

Hi I am really lost here, if you are deploying AI agents in production, how are you handling auditability? also what would make you trust an external witness protocol? fo reference, the Amazon vs Perplexisty use case!!

sshussain270 35 days ago |

Can you use public-key signatures like RSA/ECDSA like TLS certificates do but for agents entering logs? So the private key remains private to agent only, not even the LLM. Also, each log entry could include the hash of the previous one before the agent signs it so any alternation in previous log signatures break the ones after it, like block chains do.

Yahyaaa 35 days ago | |

Yes true, and this is somehow we are doing it. However, integrity is necessary but not sufficient, the chain proves nothing was changed It doesnt prove the record was accurate

sshussain270 34 days ago | | |

What do you mean by accurate? As in what was an entered by the agent a truth account of what happened or as in someone didn't come in to modify the log that agent entered?

Akaoutar 36 days ago |

I dont think that you need that. So you shoudnt be lost at all

Yahyaaa 36 days ago | |

I dont think so, You dont need independent proof until you are in court, Perplexity and Amazon both had logs, neither could prove who was behind thoses actions or who authorized what!

Akaoutar 36 days ago | | |

You may be right but I'm not convinced