White House unveils pared-back AI executive order

The White House on Tuesday released its long awaited artificial intelligence executive order, putting forth a plan that downsizes an initial version that was scrapped last month amid internal dissent.

The biggest change between the two iterations is that the latest plan changes the voluntary review period for government testing of AI models to within 30 days of release to the public and not the previously mandated 90 days. AI industry leaders had reportedly been pressuring the White House to establish a 14-day review period.

The order notes that federal access to the models should be subject to “appropriate confidentiality, cybersecurity, insider-risk, and intellectual-property protection, use, and nondisclosure requirements.”

Industry should collaborate with the government to select “trusted partners” who can access specially designated “covered frontier” models to beef up the cybersecurity of critical infrastructure and track cyber threats in a classified setting, according to the order. AI developers are directed to work with the government to decide which models to designate and which partners to entrust.

The voluntary framework also should not be seen as authorizing the “creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models,” according to the order.

The EO additionally mandates that executive branch officials — led by the Treasury Department — create an AI cybersecurity “clearinghouse” to facilitate collaboration between government, industry and critical infrastructure operators. The order directs Treasury officials to lead an executive branch effort to scan for vulnerabilities identified by AI models and determine how to prioritize patching.

It also states that the Office of the National Cyber Director, the Cybersecurity and Infrastructure Security Agency and the Office of Management and Budget should locate federal grant funding that can be used for advanced AI vulnerability detection.

The order was signed behind closed doors, a move that follows the last-minute cancellation of a public signing of the previous iteration amid internal conflict between some administration officials and former AI and crypto czar David Sacks, who told the president industry was concerned about its provisions harming innovation and competitiveness with China.

The first version of the executive order had been agreed to and approved by high-ranking administration officials and shaped with input from industry leaders like Google, OpenAI and Anthropic before it was put on hold. At the time, the president said he “didn’t like certain aspects” of the EO, citing the threat posed by China.

The president has largely directed the government to pursue an AI strategy that deemphasizes regulation, but that approach has been debated in recent weeks as new models like Anthropic’s Mythos have emerged as significant cybersecurity threats capable of discovering and targeting zero-day vulnerabilities autonomously.

Senate Intelligence Committee Chairman Mark Warner (D-VA) endorsed the provisions in the executive order. However, while he praised the various measures as necessary reforms, he also criticized the administration for tearing up a Biden artificial intelligence EO that did some of the same things.

“I salute the proposal for pre-deployment testing on a collaborative basis – just as I did when that idea was first advanced in the last administration’s EO, which was rescinded on Trump’s first day,” Warner said in a statement.