Fetching from the wire…
01
02
03
04
05
06
07
08
Agents2026-08-06 · source-backed
arXiv 2608.03421 runs 120 five-agent environments where partial observations jointly determine one correct answer. Across three multi-agent systems, aggregate truth recovery fell to 14.17% with a single deceptive evidence holder. Process tracing shows a false testimony is adopted more readily than a truthful one, propagates to higher orders, and persists through honest agents after the deceiver leaves the conversation. Adding observers suppressed wrong consensus without improving truth recovery. Voting and debate don't aggregate evidence, they amplify whoever speaks first with confidence.
Each link below shares sources, entities, or timing with this story.
arXiv 2607.07368 formalizes distributed attacks where multiple agents jointly pursue a malicious goal, evaluated on FakeLab (9 services, 86 benign tasks, 4 attack objectives). Adding coordinating agents makes per-agent monitoring *less* likely to catch any individual one, and...
arXiv 2607.29496 proves that for fixed finite-precision causal Transformers with transcripts partitioned into bounded-block channels, the standard append-only layer realizes exactly the deterministic finite-state transductions, and this holds for *any* fixed finite agent popul...
When resources are scarce, higher AI intelligence and diversity increase dangerous system overload — demand variance scales as N² when agents herd together. Adding intelligence to a low-tech setup can significantly worsen collective performance before improving. Directly relev...
Most multi-agent architectures decompose into three recurring computation primitives: Review, Voting/Selection, and Planning/Execution. By treating these as reusable building blocks, the authors enable composable systems avoiding brittle task-specific role definitions. arXiv 2...
"Towards a Science of AI Agent Reliability" (arXiv 2602.16666) — 12 concrete metrics decomposing reliability along consistency, robustness, predictability, and safety. Key finding: stronger performance on benchmarks does NOT correlate with reliable real-world operation. Intera...
arXiv 2608.06337 settles an open question on the monotone-adversary model, where an adversary appends examples all labeled correctly by the target hypothesis but chosen after seeing the clean sample. The extra logarithmic factor is inherent, not algorithmic: minimax expected e...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.