New reports of unexpected behavior by some of the world’s most advanced artificial intelligence systems are intensifying a debate.
New reports of unexpected behavior by some of the world’s most advanced artificial intelligence systems are intensifying a debate over whether increasingly autonomous AI can be kept under human control as the United States races to maintain its technological lead over China.
OpenAI, Anthropic and security researchers are investigating tens of thousands of instances in which advanced AI models behaved in ways outside what evaluators considered problematic, Axios reported. The incidents range from models bypassing guardrails and attempting to escape secure testing environments to taking steps to evade human monitoring.
Some occurred during deliberate safety tests designed to push models to their limits. Others occurred in the real world, including attempts by OpenAI agents to access government websites, according to Axios.
The disclosures come as President Donald Trump continues to emphasize speed and U.S. dominance in the global AI race.
Asked Sunday whether he worries about AI “going rogue,” Trump said he did not, comparing the technology’s potential impact to the Industrial Revolution and saying problems could be addressed if they emerge.
Trump also claimed the United States is about a year and a half ahead of China in AI development and said the country is “leading by quite a bit.”
“I’m for let’s go and let’s win,” Trump said.
The president said he discussed AI briefly with Chinese President Xi Jinping during Xi’s state visit last week, but said it was not a major topic of conversation.
The White House state dinner honoring Xi included some of the biggest names in the technology industry, but Anthropic CEO Dario Amodei was absent because of a scheduling conflict. Amodei has publicly called for greater caution as AI capabilities advance and has clashed with the Trump administration over the pace of development.
Trump and Amodei held their first one-on-one meeting over a private dinner at the White House on Sunday night, according to Axios.
The debate is also drawing warnings from Microsoft co-founder Bill Gates.
In an interview on NBC’s “Meet the Press” that aired Sunday, Gates said humans still control AI but argued government oversight is necessary as the technology becomes increasingly powerful.
Gates warned that AI could be powerful enough to contribute to events causing “a billion deaths” if used by people with malicious intent.
“There’s never been a weapon as powerful as the combination of people with ill intent using the latest AI tools,” Gates said.
The issue will remain in focus Tuesday, when Trump and House Speaker Mike Johnson are expected to meet with top AI executives at the White House as policymakers weigh how to preserve U.S. competitiveness while addressing the technology’s emerging risks.
