Blog
Notes from Marturia: cryptographic audit for AI agent decisions.
Your AI agent isn't accountable. You are.
When an autonomous agent moves money or changes a record, accountability doesn't transfer to the agent — it stays with you. The question that follows isn't 'did you log it' but 'can you prove what it
Integrity before standards: the EU AI Act's high-risk logging gap
The EU AI Act's high-risk timeline is shifting — but the logging-integrity problem underneath it isn't. Why to fix decision-record integrity now, before the standards land.
The year "we have logs" stops being enough
In 2026 the bar for AI-agent records shifts from keeping logs to proving they weren't altered. Why mutable logs fall short — and what closes the gap.
Why we don't put AI agent audit trails on a blockchain
"Put it on the blockchain" is the instinctive answer for tamper-proof AI records — and usually the wrong one. Why off-chain signed receipts fit the job better.
What the AI Act actually makes you keep
The EU AI Act tells high-risk systems to keep logs. The part teams miss: you have to be able to prove those records weren't altered. Ordinary logs can't.
LangSmith shows you what happened. It can't prove it.
LangSmith is excellent at tracing agent runs. But a vendor dashboard you control isn't evidence to someone who distrusts you. That's a different job — and the one Marturia does.
Logs aren't evidence
Observability you can edit isn't evidence. The difference between a log and proof is who can change it after the fact — and that difference decides disputes.
Why we built Marturia: signed receipts for AI agent decisions
AI agents are taking real actions but observability is mutable. We built Marturia for the moment a customer asks 'did your AI really do that?'