For companies running AI agents in production

The independent record of what your AI agents did.

When your agent acts outside its scope, you are the first party asked to prove what happened. Teracor records every action in tamper-evident form, generated by a system with no stake in the outcome.

Get in touch The problem
AB 316
California
in effect
Jan 2026
···
One defense
is gone

01: The Problem

"The AI did it" is no longer a defense.

California AB 316, in effect since January 2026, removes the argument that the AI acted on its own, for developers, vendors and deployers alike. When a claim, a regulator or a court asks what an agent actually did, most companies can only offer engineering logs written by the party whose conduct is in question.

Nothing here is legal advice. The litigation logic follows from the statute.

California
AB 316
Jan '26
In effect. Autonomous harm
defense stripped across
the deployment chain
European Union
AI Act, Article 50
Aug '26
In effect. Transparency
obligations and penalty
powers now active
European Union
AI Act, Annex III
Dec '27
High-risk obligations apply.
Deferred from Aug 2026
by the Digital Omnibus

What a log built to catch failures can see

task failed task completed in scope
in scope, task failedCrashed, and the trace caught it.
in scope, task completedWorked as intended.
out of scope
out of scope, task failedCrashed while out of scope, and caught for the crash.
out of scope, task completed

Every call returned successfully

Completed the task, acted outside its declared scope, and triggered no alarm.

A trace built to find failures records this one as a success.

02: What We're Building

A passive, tamper-evident audit trail for AI agents in production.

Teracor observes everything and blocks nothing. It records every agent action as a cryptographically verifiable receipt while leaving your agent's logic untouched. Its independence from the decision chain is structural, not a feature.

  1. 1Agent calls a tool
  2. 2Hook reports the action
  3. 3Sealed and recorded
  4. 4Checked against rules
  5. 5Verifiable receipt
One hook at tool dispatch. The model, prompts, tools and decision logic stay as they are.
The receipt that matters in court is the one generated by a system that had no stake in the outcome.

Live nowDesign partners get live API access and rule evaluation, with new capabilities as they ship.

03: What Teracor Is

Teracor is

  • A passive observation layer. It watches agent traffic and writes the record.
  • A tamper-evident record. Each receipt is sealed when recorded and can be checked by anyone.
  • One hook, no change to agent logic. Whether the agent waits for a receipt is your declared policy.

Teracor is not

  • Enforcement or governance. There is no kill switch and no blocked state. It records what agents did.
  • Developer observability. Its record is one a regulator, a carrier or a court can check without relying on the operator.
  • An insurer. Teracor is compensated identically whether a claim pays or does not.

04: Connect

Your AI agents are taking actions.
Keep a record you did not write.
Let's talk.

Teracor is looking for design partners: SaaS platforms built on foundation models first, plus teams running a third-party agent in a regulated conversation on their own license. Design partnerships start with a conversation and a sandbox under NDA, at no cost. Write with the agent you run and where it runs. Engineering teams can request an architecture walkthrough under mutual NDA.

partners@teracor.ai

Teracor is founded and led by Marc Etienne, an operator with two decades running enterprise platforms at scale. About the founder