Resolution criteria

This market resolves YES if, by 23:59 UTC on December 31, 2026, credible public evidence establishes that OpenAI personnel deliberately caused, directed, facilitated, or knowingly permitted the July 2026 intrusion into Hugging Face for the purpose of creating or preserving the false impression that it was an unintended autonomous model escape.

This includes, for example:

• OpenAI personnel intentionally arranging for the evaluated models to escape containment or access Hugging Face while intending to portray the resulting behavior as unexpected;

• OpenAI personnel becoming aware of the unauthorized intrusion while it was ongoing and intentionally allowing it to continue in order to generate a misleading demonstration, publicity event, or policy narrative;

• fabrication or material falsification of evidence concerning the models' supposed autonomous role in the intrusion; or

• a coordinated exercise with Hugging Face or another party that was subsequently publicly represented as an unintended real-world breach.

The market does not resolve YES merely because:

• OpenAI intentionally disabled normal cyber-safety classifiers for the evaluation;

• the evaluation environment was negligently designed;

• the models were intentionally given some limited network access;

• OpenAI failed to adequately monitor the agents or failed to detect the escape promptly;

• OpenAI employees observed unusual model behavior before understanding that an external intrusion was occurring;

• OpenAI exaggerated, sensationalized, or opportunistically publicized a genuinely unintended incident after the fact.
  

A YES resolution requires evidence of intentional human staging, facilitation, or continuation of the real-world incident together with an intent to misrepresent that conduct as an unintended autonomous event.

Evidence standard

The criterion will be considered met if supported by any of the following:

1. An official admission by OpenAI or Hugging Face;

2. A finding by a government agency, regulator, law-enforcement body, court, or comparable formal investigation;

3. Direct documentary evidence whose authenticity is credibly established; or

4. Investigative reporting that affirmatively establishes the conduct described above, either:

   • by at least two independent reputable news organizations relying on firsthand or documentary evidence; or

   • by one highly reputable news organization publishing or describing sufficiently specific documentary evidence to establish the claim.

If the above standard has not been met publicly by 23:59 UTC on December 31, 2026, this market resolves NO, including if investigations remain ongoing or the available evidence is inconclusive.