Secure AI Atlas mark Secure AI Atlas SECURITY & GOVERNANCE

The unresolved territory

Open Questions

This atlas records uncertainty as part of the product. Each question states what is unresolved, what knowledge it touches, and what evidence would close it.

OQ-AGENT-IDENTITY

Investigating

When does an agent stop acting as a tool and start acting as an identity?

Agent runtimes inherit or hold credentials. The model has no single place to answer who is acting. Which identity primitive should an agent hold, and who provisions it?

Evidence required

A tested identity-binding pattern with per-session attribution and revocation.

First observed: 2026-07-08

OQ-INTELLIGENCE-RELEVANCE

Open

Where is the relevance threshold for external intelligence?

The pipeline retains general security signals alongside agent-specific research. Which threshold separates material worth publishing from material worth ingesting?

Evidence required

A scoring rule validated against retained and discarded signals.

First observed: 2026-09-20

OQ-OVERRELIANCE-MAPPING

Open

How should Overreliance be mapped without collapsing into Misinformation?

Overreliance is acceptance without verification; Misinformation is false content presented as authoritative. They are related but not equivalent, and the atlas has no dedicated entry for the first.

Evidence required

Either a dedicated risk entry or an explicitly typed 'related' mapping with rationale.

First observed: 2026-09-20