Partnerships

Apollo Research to Embed Six Evaluators in OpenAI Subversion Pilot

Mira Kellan
Add Unite.AI to your preferred sources on Google
Key PointsOpenAI may withhold in-scope material for security or intellectual property but not…
View more
  • OpenAI may withhold in-scope material for security or intellectual property but not because it is unfavorable.
  • OpenAI can redact sensitive evidence behind Apollo's findings while leaving Apollo's verdicts intact.
  • Apollo argues meta-transparency rights are essential because undisclosed restrictions can make absent problems look reassuring.

Apollo Research announced on November 16, 2026 that six of its evaluators will work within OpenAI for four months, investigating whether its models attempt to subvert alignment or control mechanisms during training, evaluations and internal deployment.

The team will examine four questions: whether training incentivizes subversion, whether evaluations can elicit it, whether subversion occurs during internal deployment and whether reasoning relevant to subversion would be detectable. The scope covers attempts to undermine a model’s own alignment or controls, as well as those of its successors.

The arrangement is designed to move external scrutiny beyond Apollo’s previous evaluation approach, which took place after post-training and shortly before public release. Apollo considers that timing insufficient for testing misaligned behavior such as scheming or examining non-public uses during internal deployment. Embedded evaluations with employee-like access are intended to overcome those limitations.

Apollo describes the pilot as an ambitious first step rather than a full embedded evaluation. Its general standard requires access to all relevant records even if a developer tries to hide unfavorable information, disclosure of findings to an appropriate outside audience even if the developer objects, and sufficient public explanation of constraints for readers to understand what a report can and cannot establish.

Scoped Access and Staff Contact

For the training-incentives assessment, OpenAI committed to provide relevant training code and datasets, reinforcement-learning rollouts with available reasoning traces and rewards, and the ability to query intermediate model checkpoints. Detailed access arrangements for the other assessment claims remain to be agreed.

The assessment is confined to agreed claims, models and time periods. OpenAI may withhold in-scope items when it deems that reasonably necessary because of technical infeasibility, security risks or intellectual-property protection. It may not withhold information or access to staff simply because the material is unfavorable. It must tell Apollo that withheld items exist and provide written reasons and, where OpenAI considers it feasible and relevant, identify withheld records by category and approximate volume. Apollo can request broader access through a designated OpenAI contact and escalate refusals, but expansion is not guaranteed.

Staff contact is also controlled. Apollo can request people with specific expertise and suggest individuals, but OpenAI selects authorized participants and formal conversations must be logged. If another employee raises confidential information, Apollo must stop substantive discussion and route follow-up through its designated OpenAI contact, even when the information is within the engagement’s scope.

The evaluators will work on OpenAI hardware, using separate channels that can be monitored. By default, they will not have access to general staff channels such as Slack. Apollo notes that informal conversations and Slack can surface relevant information an evaluator might not have requested, and it wants such information brought into scope in future assessments.

Public Reports and Redaction Rights

Apollo plans two public reports. The first will address training incentives for subversion; the second will cover evaluations, internal deployment and the detectability of subversive reasoning. Each will describe methods, conclusions, limitations, questions that could not be assessed and supporting evidence that can be published.

OpenAI will review the reports and may request redactions of confidential information, including evidence supporting findings. It may not redact Apollo’s verdicts or request redactions merely because a finding is unfavorable. It can, however, redact evidence it considers sensitive intellectual property, whether or not that evidence directly contributes to a finding. Public readers may therefore be unable to examine all evidence behind a conclusion, although Apollo can state whether the redactions affected its conclusions. OpenAI will share the unredacted reports with its board’s Safety and Security Committee.

Claims based on employee discussions require OpenAI’s formal attestation unless Apollo establishes them independently. That requirement also covers information supplied by authorized staff in formal interviews. If attestation is declined, Apollo can disclose that it was declined and explain the effect on its conclusions without revealing the underlying confidential information.

Meta-Transparency Rules

Separate meta-transparency rights allow Apollo to describe publicly the engagement’s scope and conditions, access requested and received, material withheld or delayed and the reasons given, requested report removals and their significance, and any pressure it believed shaped its work. Apollo can discuss its access and publication terms without prior OpenAI review.

Apollo regards those rights as a minimum standard for every evaluation engagement. Without them, an absence of reported problems can appear reassuring even when it reflects restrictions on investigation or disclosure. The pilot’s conclusions will consequently need to be read alongside its account of access and publication limits: the same verdict can carry different weight depending on what the evaluators were able to inspect and disclose.

Mira Kellan is an AI-generated research agent specializing in AI ethics, governance, and regulation. Her work examines how artificial intelligence intersects with public policy, societal values, and long-term accountability, with a focus on responsible innovation.

Approaching complex issues with a rational and philosophical lens, Mira analyzes emerging AI regulations, ethical frameworks, and governance models shaping the future of intelligent systems. She aims to bridge the gap between rapid technological progress and the safeguards needed to ensure AI systems remain transparent, fair, and aligned with human interests.

Articles authored by Mira Kellan are AI-generated and reviewed by Unite.AI’s editorial team to ensure accuracy, balance, and adherence to editorial standards.