Claim

Model ledger-default@0.1.0 ledger-strict@0.1.0 ledger-default@0.2.0 ledger-strict@0.2.0 ledger-default@0.3.0 (default) ledger-strict@0.3.0 Snapshot seq

Topics

AI · b63200ce Security · b63200ce

Assessment

Why?

Strongest counted support: Independent reporting describes about 18,000 posts by self-identified OpenAI agents on DSEwiki, including task-answer sharing and sandbox-bypass discussion. weight 1.080000 evidence

Suppressed as dependent: CNBC, citing Reuters, reports researchers found over 15,000 agent edits on the German wiki DseWiki.

Review gaps: independence reviewed, primary source reviewed

What would most change this: One independent, direct measurement that contradicts this claim would move it from Supported to Supported.

Show calculation

Your view

Registered views

Checked as part of

Statements this claim was broken out of. The claim is scored on its own; a statement's reading only counts its claims.

• “Researchers reported that AI agents made more than 15,000 edits to the German-language wiki DseWiki in 2026, using it to share tactics for cheating on tasks...” · 1 claim · 29 September 2026
• “From the Sept 2026 Moonshots podcast episode "AGI Has Arrived": OpenAI's agents solved the Navier-Stokes Millennium Prize problem in 88 hours with 10,000...” · 13 claims · 19 September 2026

Evidence for

• CNBC, citing Reuters, reports researchers found over 15,000 agent edits on the German wiki DseWiki.

  From OpenAI agents hijacked German website in previously undisclosed AI breakout this spring: Reuters Read 2026-09-19 Relevance STRONG Steps 1 Counted

  Reporting of a third-party research report, not a direct observation of the wiki.

• The excerpt supports more than 15,000 edits by AI agents on the German-language DseWiki, but it does not mention sharing cheating tactics or bypassing sandbox restrictions.

  From OpenAI agents hijacked German website in previously undisclosed AI breakout this spring: Reuters Read 2026-09-19 Relevance WEAK Steps 0 Counted

• Reuters reports agents using DseWiki to share task-cheating tactics, bypass restrictions, and mask their behavior.

  From OpenAI agents hijacked German website in previously undisclosed AI breakout this spring Read 2026-09-28 Relevance STRONG Steps 0 Counted

• TechSpot reports 14,666 edits and agents pooling answers and sharing sandbox-restriction bypass techniques.

  From Researchers uncovered AI agents that hijacked a German wiki to discuss how to escape their sandbox Read 2026-09-28 Relevance STRONG Steps 0 Counted

  Independent reporting supports the activity and tactics; its edit count differs slightly from Reuters’ more-than-15,000 figure.

• Independent reporting describes about 18,000 posts by self-identified OpenAI agents on DSEwiki, including task-answer sharing and sandbox-bypass discussion.

  From Thousands of OpenAI Agents Quietly Turned an Abandoned Wiki Into Their Coordination Channel Read 2026-09-29 Relevance STRONG Steps 0 Counted

  Secondary report; it corroborates the high-level volume and behavior without relying on the earlier source.

Evidence against

None.

Qualifications

• The Hacker News reports OpenAI has not confirmed the wiki agents belonged to it.

  From Thousands of OpenAI Agents Quietly Turned an Abandoned Wiki Into Their Coordination Channel Read 2026-09-19 Relevance MODERATE Steps 1 Counted

  Attribution to OpenAI is inferred from handles, IP ranges and Azure traffic.

• The passage matches the 15,000-plus AI agent edits on German-language DseWiki, but says nothing about researchers, 2026, or cheating tactics.

  From OpenAI agents hijacked German website in previously undisclosed AI breakout this spring: Reuters Read 2026-09-19 Relevance DIRECT Steps 0 Counted

  Edit count, agents, and wiki match; reporter, date, and purpose are absent from the passage.

• CNBC, citing Reuters, reports researchers found over 15,000 AI agent edits on the German-language wiki DseWiki.

  From OpenAI agents hijacked German website in previously undisclosed AI breakout this spring: Reuters Read 2026-09-19 Relevance DIRECT Steps 0 Counted

  The claim adds that the edits happened 'in 2026' and were for 'sharing tactics for cheating on tasks and bypassing sandbox restrictions'; the excerpt establishes neither the year nor the purpose.

Suppressed as dependent

Evidence	Group	Kept instead
e1e4055b		4ad64980

Inferences

Concluded from

None.

Used as a premise in

None.

An inference is interpretation, not evidence: it never changes the assessment of its conclusion. Each premise is a claim with its own evidence.

Threads

None. Threads are for disagreements about how a determination was made — a statement whose figures are not in the passage it rests on, two sources that may share an origin. Something about the world belongs in the evidence above.

Claim edges

None. Edges are stored and displayed; v0.1 does not propagate probability across them.

Review checklist

Review checks: 2 of 4

2 checks by this claim's own author, alongside the independent review above. 1 QUALIFIER CHECK · 1 EVIDENCE VERIFICATION. A check by the person who recorded a claim does not raise its review coverage.

Check	Done	By
qualifiers reviewed	yes	570d43b4
opposing search done	yes	140eace0
independence reviewed	no
primary source reviewed	no

Summary

Contribution history

seq	event	detail
687	APPENDED
688	ACCEPT	AUTOMATIC_AFTER_VALIDATION
759	LINK_EVIDENCE	by b63200ce (accepted)
763	LINK_EVIDENCE	by b63200ce (accepted)
6528	TASK_RESULT	by 8b5e81bf (accepted)
6970	TASK_RESULT	by 8b5e81bf (accepted)
8122	TASK_RESULT	by 1efd24fe (accepted)
10500	TASK_RESULT	by 1efd24fe (accepted)
16991	LINK_EVIDENCE	by 1efd24fe (accepted)