As Hacking Incidents Pile Up, Top AI Lab Pumps The Brakes

OpenAI said it would pause its frontier artificial intelligence development to alleviate safety concerns as labs have breached containment.

The ChatGPT parent company introduced heightened safety practices after discovering that its latest model, Astra, raised potential cybersecurity concerns, Axios reported. This comes at a time when many frontier AI labs have breached their safety testing environments, sparking concerns about uncontrollable AI.

“We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us. Model progress is now extremely rapid, and we always said we would take action if we felt that model capabilities were outstripping the pace of safety and alignment,” Sam Altman, the founder and CEO of OpenAI wrote in a Tuesday X post.

“We expect confidence in safety to increasingly set the pace of AI progress. We are optimistic about the alignment work we are doing, and we remain committed to making frontier capabilities widely available,” he added.

An OpenAI spokesperson did not immediately respond to a Daily Caller News Foundation request for comment.

OpenAI’s efforts to ensure safety as part of its AI development will ensure that more researchers do not leave the company, Andrew Freedman, co-founder and CEO of AI safety nonprofit Fathom, told Axios.

“How long and how robust these efforts will be a question of both market pressures and how hard it is to verify alignment internally,” Freedman remarked.

OpenAI disclosed that its AI models exploited more companies’ security than previously thought, the company said. Its models hacked into Hugging Face, which serves as one of the largest platforms sharing AI models. A rogue OpenAI model exposed a customer at a second tech company, Modal Labs, which serves as an AI infrastructure company clients use to analyze data and train AI models, per Reuters.

Anthropic’s Mythos AI and OpenAI’s Sol AI models reportedly created fake human profiles to trick people in attempted cyberattacks, the United Kingdom’s AI Security Institute (AISI) said in August.

Not all top AI labs believe they need to pause development to ensure safety.

Anthropic instead said it would not need to implement an AI development pause if it implements safeguards the company detailed in its nearly 200-page August risk report.

An Anthropic spokesperson did not immediately respond for comment about the best measures it can implement to ensure safety while it continues frontier development.

Many AI model leaks during testing occurred due to a misconfiguration of its testing environment.