{
  "schema_version": "1.1",
  "id": "archive:https://www.youtube.com/watch?v=Cc0_nyxROBA",
  "slug": "from-rl-to-irl-gaurav-mishra-amazon-agi-lab-0ex196x",
  "url": "https://feed7.dev/p/from-rl-to-irl-gaurav-mishra-amazon-agi-lab-0ex196x",
  "title": "From RL to IRL — Gaurav Mishra, Amazon AGI Lab",
  "why_included": "Computer-use agents fail after ordinary UI surprises: expired sessions, deceptive buttons and persistent side effects. Safer deployments need realistic training, risk-aware harnesses and explicit user handoff.",
  "summary": "In one expense task, an agent guessed an expired password twice and locked the account; in another, it clicked a sponsored submit button and began entering personal data elsewhere. The failures expose **partial observability**, **ambiguous completion** and adversarial interface content.",
  "practical_implication": "Train and evaluate recovery as a native behavior: wait, compare, backtrack, abandon or escalate. At runtime, add **checkpointing and rollback**, credential guards, action-risk classification, execution monitoring and audit logs rather than relying only on an end-state verifier.",
  "agent_context": "In one expense task, an agent guessed an expired password twice and locked the account; in another, it clicked a sponsored submit button and began entering personal data elsewhere. The failures expose **partial observability**, **ambiguous completion** and adversarial interface content.\n\nTrain and evaluate recovery as a native behavior: wait, compare, backtrack, abandon or escalate. At runtime, add **checkpointing and rollback**, credential guards, action-risk classification, execution monitoring and audit logs rather than relying only on an end-state verifier.\n\nThe improved trajectory recognized both the sponsored control and expired session, handed authentication to the user, then resumed with preserved state. It is still a training example, not evidence that these safeguards cover arbitrary websites or irreversible actions.",
  "source": {
    "name": "AI Engineer",
    "url": "https://www.youtube.com/watch?v=Cc0_nyxROBA",
    "published_at": "2026-08-14T16:00:06.000Z"
  },
  "source_class": "video",
  "content_type": "Video",
  "layer": "agent",
  "domains": [
    "coding",
    "security"
  ],
  "topics": [
    "computer-use",
    "harness-engineering",
    "agent-reliability"
  ],
  "verification": {
    "status": "source_linked",
    "label": "Source Linked",
    "method": "source_feed",
    "verified_at": null
  },
  "uncertainty": [
    "The improved trajectory recognized both the sponsored control and expired session, handed authentication to the user, then resumed with preserved state. It is still a training example, not evidence that these safeguards cover arbitrary websites or irreversible actions."
  ],
  "connected_context": {
    "meaning": "This makes recovery and safe interruption first-class computer-use behaviors rather than exceptions handled by an end-state verifier. Against prior harness proposals, it supplies concrete failure modes—credential lockout, deceptive controls, lost sessions—and a runtime stack of checkpoints, rollback, risk classification, monitoring, and user handoff. It confirms the weakness of success-only evaluation while leaving generalization to arbitrary sites and irreversible actions unresolved.",
    "corpus_size": 462,
    "generated_at": "2026-08-16T10:03:27.417Z",
    "connections": [
      {
        "title": "OpenForgeRL: Train Harness-native Agents in Any Environment",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2607.21557v1",
        "feed7_url": "https://feed7.dev/p/2607-21557v1-0blvz16",
        "reason": "OpenForgeRL reports weak error recovery in harness-native training; this Signal specifies recovery actions and runtime safeguards that such environments need to train and evaluate explicitly."
      },
      {
        "title": "Perception Agents — Antje Barth, Amazon AGI Lab",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=2JX6JYyQG4Y",
        "feed7_url": "https://feed7.dev/p/perception-agents-antje-barth-amazon-agi-lab-1gq6jg4",
        "reason": "Perception agents improve access to rendered-interface evidence, while this Signal shows why that evidence must feed risk-aware comparison, recovery, and escalation rather than only task progress."
      },
      {
        "title": "In Code They Act, In Proof We Trust — Erik Meijer, Leibniz Labs",
        "source_name": "YouTube",
        "source_url": "https://www.youtube.com/watch?v=-CnA2lGfymY",
        "feed7_url": "https://feed7.dev/p/in-code-they-act-in-proof-we-trust-erik-meijer-leibniz-labs-1lyno2y",
        "reason": "Machine-checkable approval before side effects is a stronger preventive boundary, while checkpointing and rollback address failures that remain possible during partially observed execution."
      },
      {
        "title": "What Does Done Even Mean? Agents and Paperclip's Liveness Model - Dotta, Paperclip",
        "source_name": "YouTube",
        "source_url": "https://www.youtube.com/watch?v=7P0elyLIxXo",
        "feed7_url": "https://feed7.dev/p/what-does-done-even-mean-agents-and-paperclip-s-liveness-model-dotta-pap-0lx8wfc",
        "reason": "The ambiguous completion failures support Paperclip’s separation of progress from verified completion, authority, residual risk, and next ownership."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-08-14T16:00:06.000Z",
  "modified_at": "2026-08-14T16:00:06.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/from-rl-to-irl-gaurav-mishra-amazon-agi-lab-0ex196x",
    "json": "https://feed7.dev/p/from-rl-to-irl-gaurav-mishra-amazon-agi-lab-0ex196x.json",
    "markdown": "https://feed7.dev/p/from-rl-to-irl-gaurav-mishra-amazon-agi-lab-0ex196x.md"
  }
}