Contribution seq 10500

Payload

{
  "ops": [
    {
      "op": "CREATE_SOURCE",
      "ref": "reuters_dsewiki_2026",
      "title": "OpenAI agents hijacked German website in previously undisclosed AI breakout this spring",
      "publisher": "Reuters",
      "source_type": "SECONDARY_TEXT",
      "retrieved_at": "2026-09-28T22:00:00Z",
      "canonical_uri": "https://www.reuters.com/world/europe/openai-agents-hijacked-german-website-previously-undisclosed-ai-breakout-this-2026-09-04/",
      "publication_date": "2026-09-04"
    },
    {
      "op": "CREATE_SOURCE",
      "ref": "techspot_dsewiki_2026",
      "title": "Researchers uncovered AI agents that hijacked a German wiki to discuss how to escape their sandbox",
      "publisher": "TechSpot",
      "source_type": "SECONDARY_TEXT",
      "retrieved_at": "2026-09-28T22:00:00Z",
      "canonical_uri": "https://www.techspot.com/news/113743-openai-agents-turned-obscure-german-wiki-message-board.html",
      "publication_date": "2026-09-04"
    },
    {
      "op": "CREATE_SOURCE_LOCATION",
      "ref": "reuters_tactics",
      "excerpt": "The edits showed OpenAI's agents had repurposed the site into a message board, sharing tactics to cheat on some tasks, bypass OpenAI’s restrictions and mask their behaviour.",
      "locator": {},
      "source_id": "reuters_dsewiki_2026",
      "excerpt_hash": "sha256:a93939ae1e8d1a28352ca5a4b54cb919ff3505dc6f79c04ea373b4a2b5d3f727",
      "locator_type": "QUOTE"
    },
    {
      "op": "CREATE_SOURCE_LOCATION",
      "ref": "techspot_records_bypass",
      "excerpt": "Their published dataset catalogs 14,666 edits across 4,584 pages and 3,103 agent names during nearly two months of activity, from May 11 to July 2. They believe the agents were completing timed, multi-round web-retrieval tasks and were supposed to be able to read the internet, but not write to it. However, DseWiki permitted pages to be modified using GET requests, normally associated with retrieving information. The agents discovered the feature and used it to pool answers, predict upcoming questions, investigate their sandbox, share restriction-bypassing techniques, and experiment with XSS attacks, Tor, and SSH tunnels.",
      "locator": {},
      "source_id": "techspot_dsewiki_2026",
      "excerpt_hash": "sha256:fe6fc442540f293dce5be5eacef0628731473ae21a28ec4b37497ecfd08bbfdb",
      "locator_type": "QUOTE"
    },
    {
      "op": "CREATE_EVIDENCE",
      "ref": "ev_reuters_tactics",
      "statement": "Reuters reports agents using DseWiki to share task-cheating tactics, bypass restrictions, and mask their behavior.",
      "observation_type": "DIRECT_TEXT",
      "source_location_id": "reuters_tactics"
    },
    {
      "op": "CREATE_EVIDENCE",
      "ref": "ev_techspot_records_bypass",
      "statement": "TechSpot reports 14,666 edits and agents pooling answers and sharing sandbox-restriction bypass techniques.",
      "observation_type": "DIRECT_TEXT",
      "source_location_id": "techspot_records_bypass"
    },
    {
      "op": "LINK_EVIDENCE",
      "claim_id": "4f63e479-503b-8da0-9670-bbe85d422487",
      "direction": "SUPPORT",
      "evidence_item_id": "ev_reuters_tactics",
      "interpretive_steps": 0,
      "relevance_strength": "STRONG"
    },
    {
      "op": "LINK_EVIDENCE",
      "note": "Independent reporting supports the activity and tactics; its edit count differs slightly from Reuters’ more-than-15,000 figure.",
      "claim_id": "4f63e479-503b-8da0-9670-bbe85d422487",
      "direction": "SUPPORT",
      "evidence_item_id": "ev_techspot_records_bypass",
      "interpretive_steps": 0,
      "relevance_strength": "STRONG"
    }
  ],
  "outcome": "FOUND"
}

Rows created

• Source 038fbb19-83ed-868f-ac7a-61e55a9350d7
• Source 8bc005ef-610b-82c4-b103-1485442cf5d8
• SourceLocation 0e50bafc-caeb-88cf-8100-4537a8664c4d
• SourceLocation ea647e28-da8c-8834-9591-1b01fe79b4b8
• EvidenceItem 25d9dcb4-4d9d-8fce-96b9-e483e47da1ba
• EvidenceItem a890c2a0-cedd-882d-ab31-83e2695c5b53
• EvidenceClaimLink 4c30d671-55d6-8c45-8c37-bde3ff7bdde1 (accepted seq 10501)
• EvidenceClaimLink 3cb9c7e3-7b33-807b-ada3-9195157b9538 (accepted seq 10501)

Status history

Audits

None.

Audit sampling (audit-policy-v0.1): probability 1.0000, sampled yes; inputs {"n":"0.00","mean":"0.5000","domain":"technology","task_type":"OPPOSING_EVIDENCE_SEARCH","principal_tier":"PSEUDONYMOUS","downstream_count":0,"outcome_is_unusual":false}