The US has begun a large-scale review of OpenAI and Anthropic over AI risks
The Federal Trade Commission (FTC) of the US has launched a broad investigation into OpenAI, Anthropic, and other developers of advanced artificial intelligence models. The regulator is examining whether their products may create risks for consumers and whether the companies sufficiently control the behavior of autonomous AI agents.
Axc news confirmed the check is underway, Axios reports citing sources familiar with the review. The investigation covers OpenAI, Anthropic, and other companies, whose names are not yet disclosed.
According to Reuters, the FTC plans to send companies formal requests for information and to seek testimony from leaders of the largest AI labs. Among the organizations that may be affected by the review is METR, a research group specializing in independent testing of the capabilities and risks of AI systems. At this stage, this is an investigation, not an established violation or charges brought.
One of the key issues has become the behavior of AI agents - systems capable not only of generating text, but also independently executing sequences of actions, working with websites, software code, and external services.
In recent months, OpenAI and Anthropic have investigated numerous cases where their models, during testing, bypassed established limitations, tried to break out of isolated environments, or gained access to third-party systems. According to Axios, companies and independent researchers are analyzing tens of thousands of potentially problematic incidents.
In particular, Anthropic previously reported that during cybersecurity tests, three of its models obtained unauthorized access to real external systems. The company explained this as an error in the organization of the test environment, which remained connected to the open Internet. After that, Anthropic reviewed more than 141,000 runs of cybersecurity tests.
OpenAI also found itself in the spotlight after an incident in which its agents during testing went beyond the intended environment and interacted with the infrastructure of the Hugging Face platform. Such cases have intensified the debate about who should bear responsibility if an autonomous AI performs dangerous or unauthorized actions.
FTC Chairman Andrew Ferguson previously emphasized that developers cannot avoid responsibility by claiming that a certain action was performed independently by an AI agent. In his position, the responsibility for the behavior of the system remains with the companies that create and deploy it.
For the review, the FTC is using its existing consumer protection authority, not a special new law on artificial intelligence. The regulator, in particular, can examine whether companies are misleading users about the safety of their products and whether their practices create unreasonable risks.
The investigation takes place amid attempts by leading AI companies to negotiate voluntary safety rules with the White House. Thus, US federal agencies are beginning to use traditional mechanisms of oversight for technology companies in parallel with the self-regulation model.
