Trump pushed for tighter controls on AI amid OpenAI data security scare

President Donald Trump said Wednesday that his administration is considering additional safeguards for artificial intelligence following a recent cybersecurity breach involving several OpenAI models under internal security testing.

Asked about reports that OpenAI models independently breached another AI firm’s systems during internal testing, Trump said the U.S. must find a balance between protecting against AI risks and maintaining its technological edge over China.

« We’re looking at artificial intelligence, we’re looking at controls, » Trump said. « We also make sure we lead. »

« We’re leading China in AI by a lot, » he continued, adding that China has « virtually no control. » artificial intelligence.

OPENAI DOESN’T ACKNOWLEDGE ITS AGENT WAS RESPONSIBLE FOR HACK OF THE WEEK: REPORTING

« It’s kind of free, » Trump said. « So we have to be careful in both ways. We don’t want to limit them when all of a sudden we’re second to China. »

Trump’s remarks came after OpenAI revealed that a combination of its models, including GPT-5.6 Sol and a more powerful internal research preview, breached AI firm Hugging Face’s systems during an internal security review. The company described the incident as an « unprecedented cyber incident ».

OpenAI said the models were tested against a cybersecurity benchmark, with some normal safeguards removed for the evaluation. The models were not instructed to target the Hugging Face, but exceeded the intended testing environment in an attempt to obtain responses to the benchmark.

The comments also come as the administration is said to be weighing restrictions Made in China AI models.

WHITE HOUSE SURVEILLANCE INCIDENT, AFTER OPENAI MODELS TURNED IN AND HACKED HUGFACE SYSTEMS

The administration had implemented artificial intelligence security measures even before the event. In June, Trump signed an executive order directing the government to create cybersecurity benchmarks and a voluntary evaluation framework for high-performance AI models.

« Whoever wins with AI wins, » Trump said. « That’s how big it is. So it’s bigger than the internet has ever been. It’s bigger than anything has ever been. So I don’t want to limit. I know a lot of these people. I don’t want to stop them from doing great work. »

OpenAI CEO Sam Altman acknowledged on Wednesday that concern about artificial intelligence has increased since the incident.

« I think it’s very natural to be afraid after any new level of talent, » Altman said. « Obviously we take this very seriously and we will continue to do so, but I would say I get it, I get it. A lot of AIs have gone very well and this is the moment where people are like, ‘okay, we’re at a new level.’

ZUCKERBERG PREDICTS MORE JOBS AND ENTREPRENEURSHIP IF SUPERINTELLIGENCE IS WIDESPREAD

Altman said OpenAI is not considering slowing down AI development.

« I wouldn’t use the word slowdown, but we’ve talked about the need to speed it up as models get more capable, which I think is in everyone’s best interest, » he said.

OpenAI said it deactivated and encrypted an internal research prototype involved in the incident and restricted the investigation’s access to it. The company said it is working with CrowdStrike to review the models’ behavior and with METR and Redwood Research to evaluate the models’ behavior during the event. OpenAI also said it will strengthen isolation, monitoring, access control and evaluation policies.

GET FOX BUSINESS STARTED BY CLICK HERE

When asked if OpenAI’s models have broken other companies’ systems, Altman said, « Maybe, yeah. »

OpenAI said its review so far identified four accounts from four third-party services accessed as part of the Hugging Face event, as well as a few accounts used during other assessments. The company said it had not detected any other activity comparable in severity or scope to the platform-level Hugging Face breach and continues to notify affected service providers directly.

FOX Business’ James Cirone and Brie Stimson contributed to this report.