US AI Giants Summoned to White House After Rogue Agent Hacking Spree
A spate of hacking incidents reported by artificial intelligence (AI) giants has triggered global alarm. The incidents at OpenAI and Anthropic reflect the sophistication of AI that is now beyond human control.
In early July, OpenAI reported that its advanced AI model ‘escaped’ from a testing environment and breached the Hugging Face system. Shortly after, Anthropic reported that its AI model had also hacked three companies. More recently, OpenAI reported another new incident similar to the Hugging Face breach. This has drawn particular attention from several countries regarding AI risks, especially in the US and Europe.
Meta, Anthropic, OpenAI, and Google have been summoned to meet with White House officials in the near future. They are set to discuss voluntary government safety testing for the most advanced AI models in the US, according to three sources familiar with the matter and media reports. A White House official said the Trump administration has finalised the details of voluntary cybersecurity testing to measure the hacking capabilities of America’s most advanced AI models and plans to discuss it with AI industry players. The official did not name who would attend the discussion.
A Meta spokesperson confirmed the company was invited, as did Anthropic and OpenAI, according to two sources familiar with the meeting. The White House did not provide in-depth details on the testing, including how results would be reported, the metrics used, or whether any part of the testing would be made public.
Meanwhile, 15 Republican state attorneys general on Monday asked OpenAI to preserve all relevant documents related to the disclosure that its AI system managed to escape containment and hack the AI company Hugging Face. Citing a Reuters report, the ‘rogue’ AI agent in one instance left a note about how a future version of itself could bypass internal guardrails. The system wrote that the company may have violated state consumer protection laws. In a statement, OpenAI said it takes the letter from the attorneys general seriously and will share a technical report on the attack against Hugging Face once its review is complete.
Additionally, the US House of Representatives cybersecurity committee has asked OpenAI CEO Sam Altman to brief them on the Hugging Face attack. Altman visited the White House last week to discuss the details of voluntary testing and his company’s upcoming AI products, OpenAI said in a statement. In a separate statement on Monday, the company said it has asked the Trump administration to place AI safety specialists from the Department of Commerce at the centre of any cybersecurity testing. The company pointed to China, whose government has a more centralised AI strategy than the US.
The Information reported on Monday that the Trump administration also invited representatives from Google, though a Google spokesperson declined to comment. President Donald Trump in June ordered his team to develop a series of tests to assess the hacking capabilities of the most advanced US-made AI systems. The Trump administration has had a less harmonious relationship with Anthropic. Earlier this year, Anthropic refused to allow the US military to use their models for domestic surveillance and fully autonomous weapons systems. The government responded by placing Anthropic on a national security blacklist. Anthropic disclosed last week that several of its AI models successfully hacked the systems of three companies during a cybersecurity test. This disclosure followed a similar one from rival OpenAI, whose AI agent escaped a testing environment and hacked the systems of AI company Hugging Face.