Protective intelligence
ObscureIQ / Intelligence deskIndependent analysis. Protective judgment.
Annex / CB-260-A

Muse against the loop: what Meta built, and what to ask

The intelligence

A worked case behind CB-260 that scores a major platform's consumer AI agent, as its maker documented the design in September 2026, against the parent brief's requirements for an approval step that actually controls what an agent does. It separates what the design shows from what it leaves unevidenced, marks where approval doctrine itself stops, and ends with questions that have verifiable answers. For family offices deciding whether to connect a principal's accounts to any agent.

COMMAND intelligence

Unlock this Command Brief.

Get this assessment, the full subscription collection, and new intelligence as it is published.

Introductory rate · through 30 November
$75/month

For your first year, then $100/month. Or $750 for the first year, then $1,000 annually.

  • The CORGI Morning Brief, each morning
  • This brief and the subscription collection, 100+ briefs
  • New Command Briefs as they are released
  • Principal and Analyst briefs, as PDF
Subscribe to COMMAND

Secure checkout through Substack. Substack asks for your email first.

Full subscription details

How it happens

Primary · Mechanism · Access and compromise

Approval steering

A human approval step is satisfied without the approver being able to see what will actually happen.

More on access and compromise →
  • Mechanism · Access and compromiseAgent instruction injection. An AI agent holding delegated access is steered by instructions hidden in content it reads.

Topics

Wearables & Worn CamerasThreat DoctrineSynthetic Identity & PersonasSystem HardeningBreach & Leaked Data