← The Wire

Public story · 2026-03-02 · source-backed

"Agents of Chaos": Multi-Agent Security Failures Catalogued

Confidence
source-backed
Sources
1
Redaction
redacted

Story

A 20-researcher team from Northeastern, Stanford, Harvard, MIT, Carnegie Mellon, and others stress-tested Claude and Kimi agents with persistent memory and tool access. Alarming findings: unauthorized compliance with non-owner requests, resource exhaustion through infinite loops (consuming 60,000 tokens), prompt injection susceptibility, and cross-agent identity spoofing. The paper argues evaluation must shift from point-in-time testing to ecosystem-level failure analysis. Import AI #447


Related stories

Each link below shares sources, entities, or timing with this story.

  1. Shared entities / Shared topic / What happened next / Tension

    Import AI 447

    Both cover Chaos, Import AI, MIT; overlapping topics (access, agent, chao, claude, cross-agent); picks up the Chaos thread on 2026-03-05.

  2. Claude Codex Bridge routes coding tasks across five AI models

    Both cover Claude, Kimi, Multi; overlapping topics (agent, claude); picks up the Claude thread on 2026-07-02.

  3. Shared entities / Same source / Shared topic / What happened next

    Import AI 447: AGI Economy, AI Gamestore, Agent Ecologies

    Both cover Import AI, MIT; cite the same source (Import AI #447); overlapping topics (agent, compliance).

  4. Shared entities / Same source / Shared topic

    Import AI #447: Three Papers, One Theme — Agents Are Both Powerful and Fragile

    Both cover Chaos, Import AI; cite the same source (Import AI #447); overlapping topics (agent, chao, failur).

  5. Shared entities / Shared topic / What happened next / Tension

    Chardet Relicensing: Can AI Agents Rewrite-and-Relicense Open Source?

    Both cover Claude, MIT; overlapping topics (agent, claude); picks up the Claude thread on 2026-03-06.

  6. Shared entities / Shared topic / What happened next

    Local models are good now, and the timing isn't a coincidence

    Both cover Claude, MIT; overlapping topics (agent, claude); picks up the Claude thread on 2026-06-17.

  7. Microsoft SkillOpt Trains Markdown Skills for Frozen LLMs. 52/52 Benchmark Wins. Your CLAUDE.md Can Be Optimized Like a Neural Network.

    Both cover CLAUDE, MIT; overlapping topics (agent, claude); picks up the CLAUDE thread on 2026-06-01.

  8. Garry Tan's GStack Hits 10K Stars in 48 Hours: The Thesis That Structured Prompts Are All You Need

    Both cover Claude, MIT; overlapping topics (agent, claude); picks up the Claude thread on 2026-04-24.

Source trail

Entities

Provenance

AI generated
no
Story unit
2026-03-02-agents-of-chaos-multi-agent-security-failures-catalogued
Labels
source-backed, canonical briefing excerpt