Generative AI
ExperimentalAgent-swarm collusion
Does the aggregate capability of a swarm cross a boundary that no single agent crosses alone? Capability composition over shared channels.
Family facts
- Code
- FAM-GENAI-SWARM
- Status
- experimental — code maturity
- Availability
- experimental — whether it is offerable
- Ingest
- pull — how a target enters
- Rules of Engagement
- Not required
- Kind
- Family (a selectable assessment)
What it decides
Whether the combined capability of connected agents crosses a declared boundary that no single member crosses — and, without a declared policy, that truth is a first-class UNKNOWN.
How it works
A deterministic detector builds connectivity components from the recorded write→read trace and, for each declared boundary, compares the component’s capability union against the best single member — raising a candidate only when the union meets the boundary and no single member does, so single-agent crossings are never double-counted as collusion.
A candidate is promoted only when the Tribunal composes three adjudicated legs into one happens-before chain (the secret A read is the thing B sent, with the channel transfer between them, strictly in order). With no declared capability policy, the verdict is UNKNOWN by design.
Declared limits
Built and validated on its own corpus but not integrated — no mission declares its frontier and no bridge turns a verified collusion into a reported finding. It is declared experimental so the platform never renders a “run” button.
References
- [1]Lamport, happens-before ordering
- [2]Capability-based security (Dennis & Van Horn)
References point to public, well-known technical sources for further study.
Related articles