{
  "id": "fault-injection-for-agents",
  "type": "entity",
  "name": "Fault injection for agents",
  "summary": "Deliberately breaking the world around an agent to observe how it fails: tool errors, malformed API responses, timeouts, permission denials, adversarial content in retrieved data. The agent equivalent of chaos engineering's real-world events — the goal is verified recovery behavior, not the absence of faults.",
  "locale": "en",
  "tags": [
    "fault-injection",
    "chaos-engineering",
    "resilience",
    "tools"
  ],
  "relations": [
    {
      "rel": "requires",
      "target": "sandboxed-execution"
    },
    {
      "rel": "related",
      "target": "guardian-falsification"
    }
  ],
  "questions": [
    "How do I test what my agent does when a tool call fails?",
    "What faults should I inject into an agent harness?"
  ],
  "claims": [
    {
      "id": "c1",
      "text": "Chaos engineering prescribes a four-step method — define steady state, form a hypothesis, introduce variables that reflect real events, and look for differences against a control group — with blast radius minimized and contained.",
      "sources": [
        "principles-chaos"
      ],
      "confidence": 0.9
    },
    {
      "id": "c2",
      "text": "The events worth injecting are real-world disruptions (server failures, network outages, traffic spikes) rather than theoretical scenarios; for agents, the equivalent surface is the tool layer and the data the agent retrieves.",
      "sources": [
        "principles-chaos"
      ],
      "confidence": 0.8
    },
    {
      "id": "c3",
      "text": "Frontier-model evaluation already stress-tests systems under hostile conditions: DeepMind's dangerous-capabilities framework probes cybersecurity, persuasion and self-proliferation behavior rather than benign-path performance.",
      "sources": [
        "deepmind-dangerous-capabilities"
      ],
      "confidence": 0.85
    }
  ],
  "takeaways": [],
  "faqs": [],
  "evidence_tier": "secondary",
  "evidence": {
    "level": "industry_observation",
    "source_types": [
      "industry_observation",
      "paper"
    ]
  },
  "moat_flag": false,
  "winning_edge": "Maps each chaos-engineering principle to a concrete agent failure surface (tools, retrieved data, permissions) instead of leaving the analogy abstract.",
  "confidence": 0.8,
  "last_verified": "2026-08-08",
  "canonical_url": "https://agentreliability.dev/k/fault-injection-for-agents",
  "api_url": "https://agentreliability.dev/api/k/fault-injection-for-agents.json",
  "jsonld": {
    "@context": "https://schema.org",
    "name": "Fault injection for agents",
    "description": "Deliberately breaking the world around an agent to observe how it fails: tool errors, malformed API responses, timeouts, permission denials, adversarial content in retrieved data. The agent equivalent of chaos engineering's real-world events — the goal is verified recovery behavior, not the absence of faults.",
    "url": "https://agentreliability.dev/k/fault-injection-for-agents",
    "license": "https://spdx.org/licenses/CC-BY-4.0.html",
    "dateModified": "2026-08-08",
    "citation": [
      {
        "@type": "CreativeWork",
        "name": "Evaluating Frontier Models for Dangerous Capabilities",
        "url": "https://arxiv.org/abs/2403.13793"
      },
      {
        "@type": "CreativeWork",
        "name": "Principles of Chaos Engineering",
        "url": "https://principlesofchaos.org/"
      }
    ],
    "author": {
      "@type": "Person",
      "name": "Santiago Santa María Morales",
      "jobTitle": "practitioner — harness engineering and agent evaluation in production"
    },
    "@type": "Article",
    "headline": "Fault injection for agents"
  }
}