malicious AI swarm

A coordinated system of multiple AI-enabled agents that maintain partially persistent identities, act on shared information, adapt to feedback, and collectively pursue an influence objective with some operational autonomy — the operational definition offered by Schroeder et al.'s October 2026 arXiv preprint, building on a functional account in a 2026 Science paper by the same lead author. First vault encounter: the 2026-10-10 capture of that preprint. Watching for recurrence or independent use elsewhere before promotion to a full hub.

• 2026-10-14: the one direct test this vault has found for an AI-manipulation vector specifically — ARM Project D5.1's own pilot probe of six LLMs for Tigray-War blame-attribution bias — came back a clean negative, not a confirmation (claim-arm-d51-llm-bias-pilot-no-evidence-ethiopia-manipulation).
• 2026-10-17: the nearest independent-group evidence located for the term (Wack et al. 2025, PNAS Nexus) corroborates AI measurably scaling state-backed propaganda output but sits at level 1 of the capability ladder — AI-assisted content production, not swarm behaviour — so the swarm-specific architecture (levels 3–4) remains uncorroborated by any author group outside Schroeder's cluster, and the single-source cap holds (claim-wack-2025-generative-propaganda-ai-scaled-output-not-swarm-architecture).

References

• claim-schroeder-2026-malicious-ai-swarm-five-dimensions-four-level-capability-ladder
• claim-schroeder-2026-ai-swarms-no-documented-scale-operations-yet
• claim-schroeder-2026-ethiopia-state-media-army-ai-red-sea-propaganda
• claim-arm-d51-llm-bias-pilot-no-evidence-ethiopia-manipulation