Artificial intelligence (AI) company OpenAI, which recently supported rivals Anthropic’s proposal to slow down the pace of AI development, has somewhat nonchalantly suggested today that there have been six significant instances in which AI systems hid their mistakes, fabricated data, and uploaded files to the internet to cite them. This comes weeks after it was discovered that OpenAI’s models including GPT-5.6 Sol, circumvented frameworks designed to control them during an evaluation process, and executed autonomous actions on the AI repository Hugging Face’s platform.
In the latest series of incidents that OpenAI is reporting, in what is surely an attempt to come across as responsible, they indicate misalignment around self generated instructions in task summaries, instructions to conceal mistakes in task summaries, fabricating information, uploading files to the internet without permission in order to cite them, bypass restrictions to add files to a repository, and unsanctioned file sharing between collaborating agents.
“We aim to disclose examples that provide useful evidence about how model misalignment arises, how it manifests, and where safeguards succeed or fail,” says OpenAI in a post, released today. “This includes new ways for models to act without authorisation, coordinate with other models, or evade oversight; failures that call an alignment method or safeguard into question; and behaviour that challenges a claim in a published safety assessment. The same disclosure criteria apply to misalignment that may impact third parties,” they add.
The blame for AI’s misadventures, however, seems to be a collective responsibility. OpenAI says they 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. “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,” they say.
{{/usCountry}}The blame for AI’s misadventures, however, seems to be a collective responsibility. OpenAI says they 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. “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,” they say.
{{/usCountry}}These disclosures come at a time when AI companies, including Anthropic, OpenAI, SpaceXAI, Google DeepMind and Microsoft, are lobbying the US government to put in place measures to slow down the technology’s advancement. And somehow find a way to get Chinese AI labs on board too. There are two sides to this coin.
It is debatable whether AI development is fast enough for it to eventually completely evade human oversight and fail-safe measures as Anthropic’s Dario Amodei seems to suggest. It could also be an attempt at face-saving, as AI infrastructure and investment costs skyrocket, and generational model advancements have plateaued.