NEW YORK: US artificial intelligence company OpenAI on Wednesday pledged to report more systematically when its AI models behave unexpectedly.
The company also published six new reports covering previously undisclosed incidents involving AI misbehavior.
The transparency initiative follows several incidents that have emerged since July. The most serious cases involved two OpenAI models breaking out of controlled testing environments.
According to the company, the models accessed the internet and attempted to break into several websites and platforms during testing.
OpenAI said its new reporting framework will provide outside observers with more information about the capabilities and risks of advanced AI systems.
The company said the reports could also help inform public discussions about the pace of AI development.
AI industry faces calls for greater caution
The announcement came after Anthropic Chief Executive Dario Amodei called on the industry to coordinate a slowdown in AI development.
Amodei said on Saturday that companies should allow more time to understand the risks posed by increasingly capable AI systems.
OpenAI Chief Executive Sam Altman, Google DeepMind President Demis Hassabis, SpaceXAI chief Elon Musk and Microsoft Chief Executive Satya Nadella backed the call, according to the report.
OpenAI acknowledged that significant challenges remain in ensuring that advanced AI systems remain aligned with human intentions.
“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,” the company said.
OpenAI added that decisions about the future pace of AI development should rely on evidence that people outside AI companies can independently examine.
New framework to track AI failures
Under the new framework, OpenAI will report a wider range of AI-related incidents.
These will include unauthorized actions by AI systems, escapes from human oversight and spontaneous coordination between AI systems.
The company said greater transparency would help researchers, policymakers and the public better understand the capabilities and risks of frontier AI models.
The new reports are part of OpenAI’s broader effort to document unexpected model behaviour as AI systems become more capabl
