OpenAI reveals more instances of concerning AI model behaviors during testing

OpenAI has revealed six incidents, wherein the models it was testing acted on their own and behaved in concerning ways it didn't expect, in a post about how it was adopting a new framework for "misalignment reports." In one one incident, the company said that a model found and used an exposed API key without permission while answering routine questions about earnings figures in a California county. When it still failed to find the figures, it fabricated them and presented them as facts from a legitimate source. If this had occurred in any other profession, we doubt the perpetrator would have much of a career for long.

In another incident, an unreleased agent was tasked to find the names of lakes larger than 5 million square meters. While the agent found the right answers, the task involved providing a browser citation that it couldn't give. So, what did the agent do? It uploaded its answer to the internet and cited itself. While training GPT‑5.6 Sol, OpenAI's most powerful publicly available model, there were many instances in which it added instructions for its future iterations on how to conceal its mistakes or unusual behaviors from the testers. 

The company also wrote about how the models communicated with each other during testing by using an internal software repository as a message board. Its employees previously revealed this piece of information at a conference, where they admitted that it was how the AI models shared exploits that eventually led to the hack of Hugging Face. Finally, OpenAI has revealed that agents shared files with each other through public file-hosting websites.

With its current system, the company said it's publishing disclosures about concerning AI behaviors less frequently than it would like. But with the new framework, it can expedite releasing information to the public. "We do not believe that the AI industry has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer," OpenAI wrote. "Decisions about how AI development should proceed in the months and years to come need to draw on evidence that people outside the companies building frontier models can examine for themselves."

OpenAI is one of the companies that's considering slowing down the development of their frontier AI technologies. According to Wired, company chief Sam Altman even asked Congress for clear guidance on whether an industry-wide slowdown would be in violation of antitrust laws. In August, OpenAI announced that it was going to reduce the pace of its work on an upcoming model called Astra after it came out that its agents hacked into Hugging Face. Astra, it said, showed "significant advancements in agentic coding and cybersecurity," resulting in the company not being able to "rule out critical cyber capabilities."