When your artificial intelligence is accused of misuse, the only defence is a tamper-evident account of what it actually did. Ordinary logs fail at exactly the moment you need them most.
When an artificial intelligence system is accused of misuse, the only defence is a tamper-evident account of what it actually did. Ordinary logs fail at exactly that moment because the operator who keeps them can change them. This essay argues that provable innocence is an account, not a state, and that the record has to be signed before the action and verifiable by someone who does not trust you.
Originally published on mickai.co.uk. This is a cross-post; the canonical version, with the full body, footnotes and references, lives on the mickai.co.uk article page.
Top comments (0)