Blog

Agent accountability, written down.

Verifiable receipts, ground-truth verification, and the regulation arriving for AI agents.

1 September 2026

The EU AI Liability Directive is dead: what actually governs AI agent harm now

Withdrawn in October 2025 after two years without agreement. What the revised Product Liability Directive covers instead, and the gap it leaves for AI agents used business-to-business.

25 August 2026

Indirect prompt injection: how an AI agent gets hijacked by the data it reads

Instructions hidden in the content an agent reads, why OWASP says there is no fool-proof prevention, and what to record so a compromised agent’s actions stay provable afterwards.

18 August 2026

Transparency log witnesses: how cosigned checkpoints make a log tamper-evident

What a witness actually checks, why a split-view attack defeats a self-hosted Merkle tree, the obligations you accept when a third party pins your checkpoint — and how to verify ours, which gained an independent witness on 17 August.

11 August 2026

AI agent audit trails: what to record, and what your records can prove

The fields an agent audit trail needs, the EU AI Act's six-month log retention floor in Articles 12, 19 and 26, and the gap between a record that is complete and one a third party can believe.

4 August 2026

How to sandbox an AI coding agent (and what a sandbox cannot prove)

A working bubblewrap profile, the DNS leak we found when we actually ran it, and the point at which a containment boundary stops being something you can demonstrate to anyone else.

30 July 2026

Alibi: grade coding agents on what actually happened on disk

A small open-source harness that runs the same missions through agentic CLIs and grades every run against ground truth on disk, never the transcript — with runs sealed by signed receipts a witness can cosign.

28 July 2026

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

What evidence actually demands, where centralised logging, WORM storage and hash chains each stop, and what a canonical signature over an execution record adds.

21 July 2026

How to verify what an AI agent actually did

A practical walkthrough of traceseal-verify and the anatomy of a signed execution receipt: what the signature covers, what an [OK] proves, and what it honestly does not.

19 July 2026

EU AI Act Article 50: what it means for teams running AI agents

Transparency obligations apply from 2 August 2026. What Article 50 requires, who it covers, and why signed execution receipts turn compliance into a one-command check.