
For companies with AI agents and automations in use
Every AI action checked before it runs, and kept on a record.
Your agents send emails, issue refunds, book jobs and change records. When one does something nobody approved, the first question is what it did and under whose rules. Most companies can't answer it.
Who asks, and for what
Someone already wants this proof.
The problem
What happens today
Agents act with broad access, and their logs sit inside the same tools they act through. The record belongs to the thing being watched. When something goes wrong, you rebuild what happened from fragments.
Why it matters
They decide with money
The rules are arriving faster than the tools. Europe will require logs for high-risk AI from December 2027, and from December 2026 its product liability law lets courts presume a defect when a company can't produce the evidence. Insurers have started writing AI liability cover, and they will ask what you can show.
The risks
What it costs when you can't show it.
Share of organizations in one survey that can trace an agent's action back to the human who sponsors it. 21% keep a real-time agent inventory.[1]
Share of agentic AI projects Gartner expects to be canceled by the end of 2027.[2]
From December 2, 2027, the minimum time businesses using high-risk AI in the EU must keep its automatic logs. Fines under the Act reach EUR 35 million or 7% of worldwide turnover.[3]
EU product liability rules apply to software. A company that can't disclose the evidence can have a defect presumed against it.[4]
If nothing changes
It gets bigger every year
Each new agent adds actions nobody reviews. Without one record, every incident is a forensic project, and every customer security review is a questionnaire you answer from memory.
What Piehl does
The proof, kept for you.
Rules every agent works inside
Each agent works inside the rules you set, with a named person who answers for it. Anything outside them is stopped and kept as proof.
A person on the money
Payments, refunds, signatures and deletes always wait for a named person, who sees the exact request before approving it once.
Every action on the record
What each agent did, when, under which of your rules, and who answers for it.
A record outside the agent
Tamper-evident and kept apart from the tools the agent uses. Ready for an auditor any time.
How it starts
Four steps.
- 1We list every agent and what it can touch.
- 2We set the rules for each agent and name the person who answers for it.
- 3From then on every action is checked before it runs.
- 4Auditors, insurers and customers get the export or a link.
Book 15 minutes
See your version of it.
Bring what you have to prove or what you need to see. No price talk on the first call.
Sources
- Cloud Security Alliance survey on non-human identities, 2026.
- Gartner press release, June 25, 2025.
- EU AI Act, Regulation (EU) 2024/1689, Articles 12, 26(6) and 99, with high-risk dates moved by the Digital Omnibus, Regulation (EU) 2026/1744 (Annex III from December 2, 2027; Annex I from August 2, 2028), per Gibson Dunn and Sidley summaries.
- EU Product Liability Directive, Directive (EU) 2024/2853, applies from December 9, 2026.
Figures marked vendor come from companies that sell in the market. Nothing on this page describes a Piehl customer result.