With Donald Trump, you can never be sure he won’t change his mind – or be talked into doing so – in basically a minute. Still, the White House’s apparent shift in AI policy approach is causing waves in the tech industry.

On Tuesday, the scene was rocked by the news that Alphabet’s Google, Microsoft, and xAI have reached an agreement with the Trump administration to share early versions of their AI models to assess their capabilities and security before releasing them to the public.

Of course, ChatGPT maker OpenAI and Claude owner Anthropic had already been voluntarily working with the US Center for AI Standards and Innovation (CAISI) – a team of US government scientists – to test unreleased models for vulnerabilities.

But this is now looking more ambitious and centrally-directed, experts and industry observers say.

“We can’t stop it with foolish rules”

Plainly speaking, the new agreements allow CAISI to evaluate existing and upcoming AI models and their potential impact on national security and public safety ahead of their launch.

“Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications,” CAISI Director Chris Fall said in a statement.

“These expanded industry collaborations help us scale our work in the public interest at a critical moment.”

In a way, the development isn’t that surprising. Anthropic’s powerful new Mythos AI model has pushed concerns about the tech’s impact on cybersecurity to a tipping point, and the company even said it doesn’t feel comfortable releasing the model publicly yet.

However, The New York Times now reports that the White House is even considering a slate of executive actions to address escalating security risks from advanced AI models. Apparently, it’d be a very formal review process.

To sum up, it looks like the Trump administration is ditching its noninterventionist “hands-off” approach and shifting its position on regulating AI development.

In other words, Trump is behaving a lot like the previous US President Joe Biden wanted the government to – even though, once inaugurated, he quickly rolled back a Biden administration regulatory process that asked AI developers to perform safety evaluations.

“We have to grow that baby and let that baby thrive. We can’t stop it with foolish rules and even stupid rules,”

Donald Trump.

“We’re going to make this industry [AI] absolutely the top, because right now it’s a beautiful baby that’s born,” Trump said in July.

“We have to grow that baby and let that baby thrive. We can’t stop it with foolish rules and even stupid rules.”

Last year, the US President also signed a “sweeping Executive Order to fast-track federal permitting, streamline reviews, and do everything possible to expedite construction of all major AI Infrastructure projects.”

Shadows of doubt over tech industry

To many, it’s pretty obvious that Trump’s stance has been heavily influenced by energetic laissez-faire venture capitalists like David Sacks and Marc Andreessen.

The latter two aren’t commenting. But the wider tech industry is certainly worried now, judging by comments from Daniel Castro, president of the Information Technology and Innovation Foundation, a think tank supported by Anthropic, Microsoft, and other tech companies.

“This is a full embrace of the precautionary principle. It would mean firms need government permission to innovate. That flips the default from building freely to asking first,” Castro raged on X.

“It would politicize AI. Every company would have to bend the knee to each administration – or risk losing access to the market,” he added before warning that the US needs to be moving quickly if it wants to compete with China.

The cybersecurity folks are also worried. Muhammad Yahya Patel, vCISO and cybersecurity advisor for Europe, the Middle East, and Africa at Huntress, told Cybernews that while this move is a win for US national security, it also created a massive trust gap for the tech industry.

“The industry is effectively trading a degree of autonomy for a license to operate in a highly scrutinized geopolitical environment. There is a risk that this review process becomes a tool for picking winners and losers in the AI race based on their willingness to comply with government demands,” said Patel.

He further warned that by centralizing the blueprints – if, of course, that’s what’s going to happen – of the world’s most powerful AI models in one government agency, the White House would create massive risks.

“If a government server is breached, an adversary doesn’t just get an AI model. They get the crown jewels of AI innovation from the tech giants,” Patel pointed out.

Small-scale testing won’t work

Other cyber pros who spoke with Cybernews are actually quite happy. For instance, Charlotte Wilson, head of enterprise at Check Point, said it’s logical for the US government to react to the furor around Mythos and competing models.

“It is encouraging that governments are recognizing the need to vet technology for cybersecurity concerns before release, given that taxpayers often foot the bill for large-scale breaches,” said Wilson.

Saif Khan, a fellow at the Institute for Progress think tank, told Politico that the emergence of Mythos has quickly changed the conversation around AI and national security in the White House.

“The UK AI Safety Institute is using AI cyber ranges that mirror the scale and complexity of real-world environments. Any US oversight should include similar capabilities,”

Jared Atkinson.

“The pure, Silicon Valley venture-capital type of approach to AI policy just might be over in the Trump administration,” said Khan.

Ronald Lewis, head of cybersecurity governance at Black Duck, told Cybernews that the Trump admin’s move could translate into a win for AI companies, which have so far been distrusted by a large number of Americans.

A Pew Research Center poll last November found that 50% of Republicans and 51% of Democrats said they were more concerned than excited about the increased use of AI in daily life.

“The strategic decision to introduce emerging models to CAISI therefore serves a dual purpose: it signals responsibility and cooperation with the government, while simultaneously stimulating demand in a security marketplace where fear, uncertainty, and complexity have always been powerful commercial drivers,” said Lewis.

Finally, though, it’s smart to wait and see precisely what kind of vetting of those frontier AI models the Trump admin will introduce.

Jared Atkinson – the CTO at SpecterOps, a company that worked with the United Kingdom’s AI Security Initiative to build “The Last Ones,” a cyber range used for advanced, end-to-end security testing, particularly for evaluating AI models – told Cybernews that small-scale testing environments won’t work.

The conversation on this topic is live. Join in the discussion.

“They artificially constrain both AI models and what red teamers and defenders can do in testing their capabilities,” said Atkinson.

“The UK AI Safety Institute is using AI cyber ranges that mirror the scale and complexity of real-world environments. Any US oversight should include similar capabilities.”

---

Unlock more exclusive Cybernews content on YouTube.