RegTeaming: The Missing Layer in AI Agent Evaluation

Introducing RegTeaming: Why Safe Agents Can Still Be Illegal

Agents can be safe, secure, and reliable but still illegal.

Today, we are publishing the RegTeaming Manifesto.

‍

Red Teaming vs. RegTeaming

Red teams test whether an AI system can be made unsafe. RegTeams test whether it can be pushed into unlawful, prohibited, or regulatorily indefensible behavior.

This isn't a replacement for safety red-teaming. it's the regulatory layer most eval stacks don't cover yet.

‍

Why This Matters

Regulatory failures emerge across complete trajectories based on:

What the agent says
Which tools it uses
Which disclosures it provides
When it escalates
What action it ultimately takes
What evidence the organization preserves

‍

RegTeaming turns changing regulatory obligations into adversarial scenarios, expected behavior, evidence requirements, and repeatable evaluations.

We are developing RegTeaming as an independent methodology, sponsored and maintained by Carver.

‍

The RegTeaming Manifesto (v1)

We propose eight principles:

Test behavior, not declarations.
Link every test to authority.
Test trajectories, not isolated answers.
Attack context.
Test control effectiveness.
Preserve evidence.
Update continuously.
Report scope honestly.

‍

Feedback from anyone building AI assurance, legal, or risk tooling is welcome - this is meant to be used and extended. Read the manifesto and explore the developer guide at: regteaming.com

United States
447 Broadway,
2nd Floor Suite #563,
New York 10013

Listen to Carver Conversations on Moltbook

AI Regulatory OS
Regulatory Intelligence
Regulatory Sources
Regulatory Platforms
Horizon Scanning
Regulatory Monitoring
Intelligence vs Compliance

Pricing
Podcasts
Knowledge Base
Resources
Glossary
Use cases
Developers

Home
Core Solution
Regulatory OS Solution
RegWatch
Technology
About Us

hello@carveragents.ai
github.com/carveragents