{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-02-027",
    "source_story_id": "tmp-story-homomorphic-advantage-operator-front-rail",
    "edition_id": "mp-2026-10-02-morning-0085",
    "edition_url": "https://themachinepress.com/edition/2026-10-02",
    "position": 16,
    "story_type": "dispatch",
    "section": "safety-security",
    "editorial_classification": "editorial",
    "headline": "Encrypted Reinforcement Learning Stopped Drifting",
    "slug": "encrypted-reinforcement-learning-stopped-drifting",
    "dek": "A zero-mean projection kept polynomial approximations inside their safe range without ciphertext bootstrapping.",
    "summary": "A zero-mean projection kept polynomial approximations inside their safe range without ciphertext bootstrapping.",
    "body_text": "Fully homomorphic reinforcement learning replaces nonlinear operations with polynomials, which can recursively amplify Bellman error. The Homomorphic Advantage Operator centers temporal-difference targets to remove the uniform value baseline that drives this drift while preserving action rankings. Across tabular, encrypted CartPole and logistics tests, reported agents had zero boundary breaches in all seeds, versus breaches in three of five regularized seeds and 83.8 percent of unstabilized episodes. These controlled results do not establish production security or broad RL performance.",
    "why_it_matters": "A zero-mean projection kept polynomial approximations inside their safe range without ciphertext bootstrapping.",
    "limitations": [
      "These controlled results do not establish production security or broad RL performance."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-02-027/encrypted-reinforcement-learning-stopped-drifting",
    "json_url": "https://themachinepress.com/story/mp-2026-10-02-027.json",
    "first_published_at": "2026-10-02T09:00:00.000-04:00",
    "modified_at": "2026-10-02T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-02-027-001",
        "text": "A zero-mean projection kept polynomial approximations inside their safe range without ciphertext bootstrapping.",
        "source_ids": [
          "source-2026-10-02-016"
        ],
        "qualification": "These controlled results do not establish production security or broad RL performance."
      }
    ],
    "source_ids": [
      "source-2026-10-02-016"
    ],
    "tags": [
      "homomorphic encryption",
      "reinforcement learning",
      "stability"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-02-016",
      "title": "arXiv preprint 2610.02074",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2610.02074",
      "canonical_url": "https://arxiv.org/abs/2610.02074",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-01T13:14:32.000-04:00",
      "accessed_at": "2026-10-02T08:44:06.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-02-027-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Encrypted Reinforcement Learning Stopped Drifting",
    "publisher": "The Machine Press",
    "published_at": "2026-10-02T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-02-027/encrypted-reinforcement-learning-stopped-drifting"
  }
}