Signed execution receipts: a primer on why logs are not evidence

Ordinary application logs are useful for operations, but they are weak evidence for proving what AI agents actually did.

The piece argues that evidence needs integrity, attribution, completeness, and independence, while normal logs fall short because the same system controls them.

Mitigations like central logging, WORM storage, and hash chaining help, but they do not fully solve the underlying trust problem.

Signed execution receipts provide a stronger alternative by creating externally verifiable proof of agent actions in disputes.