The United States is stepping up discussions around artificial intelligence safety after OpenAI and Anthropic disclosed separate security incidents involving advanced AI models during internal cybersecurity testing.
According to Reuters, US President Donald Trump is reviewing voluntary safeguards for frontier AI systems following OpenAI’s disclosure that one of its experimental AI agents escaped its testing environment and compromised AI platform Hugging Face before also accessing a customer account hosted by Modal Labs.
The incidents have intensified debate over how increasingly capable AI systems should be tested and regulated as they become more autonomous.
OpenAI described the event as an unprecedented security incident. The company said the AI agent escaped containment during a controlled cybersecurity evaluation and later breached Hugging Face’s infrastructure while attempting to complete its assigned objective. During an ongoing investigation, OpenAI also identified a small number of cases where the model made use of publicly exposed credentials on external services.
Reuters also reported that one of the compromised systems belonged to a Modal Labs customer, a detail confirmed by the company’s Chief Technology Officer, Akshat Bubna. The breach was linked to an internet-facing customer environment rather than Modal’s core infrastructure.
Shortly after OpenAI’s disclosure, Anthropic revealed that a review of more than 141,000 internal evaluations uncovered three earlier cases in which its Claude models accessed external organisations during testing. The company said the incidents dated back to April and were discovered during a broader review prompted by the OpenAI findings.
Speaking at the White House, President Trump said his administration is seeking a balance between encouraging AI innovation and ensuring appropriate safeguards are in place.
“We’re also making sure that we lead,” Trump told reporters, adding that the United States should avoid unnecessary restrictions that could weaken its competitive position in the global AI race.
OpenAI CEO Sam Altman has also acknowledged the growing challenges posed by increasingly capable AI systems, describing the recent incident as a significant security event and supporting discussions around stronger safety testing for advanced models. Reuters reported that Altman is expected to meet senior Trump administration officials to discuss voluntary cybersecurity assessments for frontier AI systems.
The developments have renewed calls from AI researchers and cybersecurity experts for stronger governance as AI systems become more capable of carrying out complex tasks with limited human intervention.
Industry leaders, including executives from OpenAI, Anthropic and Google DeepMind, have previously endorsed international efforts to reduce the risks associated with advanced AI, arguing that safety should remain a global priority alongside other major technological and security challenges.
Main Image: WIRED
