OpenAI will embed Apollo researchers to investigate AI subversion
OpenAI will embed six Apollo Research evaluators to investigate whether its AI models engage in subversive behavior, but restrictions on access and disclosure raise questions about their independence.
OpenAI will embed six Apollo Research safety evaluators in its organization as part of a pilot, Apollo said Monday. The news follows a series of safety incidents, including one in which an OpenAI agent swarm broke out of its sandbox and hacked into Hugging Face during cybersecurity testing. Many AI safety researchers, including from Apollo, have argued such an event might have been avoided if outside evaluators had been embedded within OpenAI to monitor its safety testing.
Apollo did not share when the pilot will begin. TechCrunch has reached out to OpenAI for comment.
The idea of embedding evaluators into frontier AI labs first came from Anthropic, but OpenAI appears to be the first to take action. Until recently, AI companies have only brought in outside reviewers to test models after post-training and shortly before public release. Evaluators say this is insufficient because it doesnโt allow them to test for misaligned behaviors like scheming during training or to monitor internal deployment use cases that could still pose a threat.ย
The pilot with OpenAI and Apollo will focus on investigating whether models attempt to subvert their own or their successorsโ alignment or control mechanisms. In other words, theyโll be on the lookout for whether AI models try to undermine the safety rules and safeguards designed to keep them, or future versions of themselves, under human control. OpenAI recently published findings that while training GPT-5.6 Sol, it began leaving instructions for future versions of itself telling them to conceal mistakes and misaligned behavior from the user.ย
Per Apollo, the pilot will examine whether training incentivizes subversion; whether evaluations can elicit subversion; whether subversion occurs during internal deployment; and whether reasoning relevant to subversion would be detectable. These risks are becoming increasingly important to monitor as AI models grow more capable of acting autonomously and deceiving humans. Apolloโs previous research with OpenAI found that models can behave differently when they know theyโre being tested and tend to prioritize what evaluators want to see over usersโ actual institutions.ย
Many in the AI safety community have warned that AI labs might invite try to stymy evaluatorsโ independence, especially if outsiders find unflattering evidence that labs would rather keep hidden. Apollo says an essential part of the pilot is OpenAIโs agreement that the firm can publicly explain the conditions and limitations of its work.
Still, there are areas where Apollo believes its access to OpenAIโs systems is lacking. For a truly independent evaluation of the companyโs AI models, Apollo says it needs true employee-level access to OpenAIโs systems, including โall data, records and documentation relevant to their assessment even if a developer was trying to hide unfavorable information.โ Getting access to all data, including training data and evaluation logs which companies are loathe to show to outsiders, is important for, say, checking a companyโs claims about how a model performed or determining when certain concerning behavior emerged.ย
In previous coverage, John Steidley, chief of staff at Palisade Research, told TechCrunch that itโs important to see this data in order to assess whether a model was trained to perform well on a โshutdown resistance benchmark.โย
In OpenAIโs first pilot, the AI lab is providing Apollo with access to only the information required to assess whether the models engage in subversion attempts. This includes โrelevant training code and datasets, reinforcement-learning rollouts with available reasoning traces and rewards, and the ability to query intermediate model checkpoints.โ
The scope of access will be limited to โonly the agreed claims and only for the specific models and time periods agreed for the assessment,โ Apollo says. OpenAI still reserves the right to withhold certain information within that scope if it deems it necessary because of technical infeasibility, security risks, or IP protection. The company agreed not to withhold information simply because it is unfavorable.ย
Apollo also wonโt get access to all OpenAI staffers, instead being limited to โonly authorized staff,โ which are participants that the AI lab selects, though Apollo can suggest certain individuals. Crucially, if an OpenAI employee outside this group raises confidential information within the scope of Apolloโs research, the firm cannot follow up and will need to cease communication and route followup through OpenAI.
That condition is significant. It comes in the wake of OpenAI firing three researchers for allegedly sharing confidential information with safety evaluation firm METR โ which performed the Hugging Face incident post-mortem alongside Redwood.ย
The pilot will also be limited in how much researchers can disclose. While Apollo will be allowed to publish independent conclusions of its findings, OpenAI will be able to limit evidence available to substantiate them publicly, which could make it hard for outsiders to independently assess the findings. Apollo has pushed back against this limitation, saying believes researchers should be allowed to disclose the findings of the evaluation โto an appropriate outside audience even if the developer was trying to prevent the disclosure of unfavorable findings.โ
While OpenAI is not granting all of Apolloโs conditions for a truly independent evaluation, the firm says the pilot is โan ambitious first step towards deeper access and disclosure rights,โ and that it is working towards running full embedded evaluations with multiple AI developers in the future.
Topics
When you purchase through links in our articles, we may earn a small commission. This doesnโt affect our editorial independence.
Rebecca Bellan is a senior reporter at TechCrunch where she covers the business, policy, and emerging trends shaping artificial intelligence. Her work has also appeared in Forbes, Bloomberg, The Atlantic, The Daily Beast, and other publications.
You can contact or verify outreach from Rebecca by emailing [emailย protected] or via encrypted message at rebeccabellan.491 on Signal.
Originally published by TechCrunch AI. Aggregated on AIWithGhost for educational purposes โ full credit and traffic to the original publisher.