Trump's AI Stance and Anthropic's Safety Paradox

Hello, and welcome to TechScape. I’m your host, Blake Montgomery, the US tech editor at the Guardian. Today we’re discussing Donald Trump’s neediness for AI and the contradictions of Anthropic’s safety-first posture.

Don’t kid yourself, Trump is not going to regulate AI

Any prediction that Donald Trump will rein in the rapid development of artificial intelligence for safety reasons is laughable at this point.

He issued a watered-down executive order on model review last week and another demanding the US military accelerate AI adoption.

Rather than restrain growth, he’s discussing buying stock.

Trump said last week that his administration would “look into” taking financial stakes in the US’s leading artificial intelligence companies. Sam Altman has reportedly participated in discussions of such stock purchases with senior White House officials, indicating the discussions are somewhat serious.

In typical vague but confirmatory fashion, Trump told reporters on Friday: “There’s something very interesting about it, where it almost becomes a partnership with the American public. We’ll ​look into that.”

A US stake in AI companies could cut two ways. Trump could use the government’s leverage as a major shareholder to restrict AI development and align it with safety incentives more than financial ones.

Or he could encourage AI firms to grow as lucrative and large as possible so the federal government can cash out like a venture capital firm. My money is on the latter.

Should Trump push the growth approach, it does not seem likely he would press AI startups in the US to slow things down in the name of safety, whether in the form of constraining models’ capabilities or halting the construction of hyperscale datacenters. He’ll want his money’s worth.

Two of Trump’s executive orders issued last week likewise point to a continuation of his growth-at-all-costs approach. One executive order sought but did not mandate a government review of AI models 30 days before their release.

He had postponed signing the order in late May, telling reporters: “We’re leading China, we’re leading everybody, and I don’t want to do anything that’s gonna get in the way of that lead.” The previous version of the order reportedly would have enforced stricter standards: mandatory review, 90 days in advance.

Trump’s second executive order directed the defense department to accelerate AI adoption, particularly with regards to national cybersecurity. In it, he put into writing the now obvious core aspect of his stance on AI, asserting that the US leads in AI “because we refuse to stifle this innovation with overly burdensome regulation”.

Don’t expect his posture to change any time soon, and don’t look to him for a check on the headlong expansion of the US’s AI giants. The only check or balance we seem to have on that is Anthropic’s own conscience.

OpenAI filed to go public on the US stock market

OpenAI confidentially files for initial public offering on US stock market

Apple’s annual developer conference

Apple debuts revamped ‘Siri AI’ and new child safety features for iPhones and iPads

Tech and power

The Guardian view on children and the internet: rolling back big tech’s untrammelled power | Editorial

Silicon Valley including Meta has embraced Maga politics, says Nick Clegg

Bernie Sanders’ AI sovereign wealth fund plan is good. But we think this is better | Nathan E Sanders and Bruce Schneier

A pause on AI development would be very good business for Anthropic

Anthropic said the world should have the option to “pause” AI development.

It also filed to go public on the US stock market.

A freeze would be very good for the company, which is leading the AI race.

Last week, Anthropic advocated for a possible “temporary pause” on advancing the capabilities of AI, saying it would convene policymakers to discuss the risks of the cutting-edge technology. In the same post, the company said its progress developing its chatbot Claude would someday soon lead to “recursive self-improvement”, a model with the capability to make better and more powerful versions of itself.

A few days prior, Anthropic confidentially filed for an initial public offering on the US stock market, cementing its status as more valuable than OpenAI for now. A freeze on AI development as the company, valued at some $965bn, overtakes OpenAI, its main rival in the AI race, worth roughly $850bn, would be quite convenient.

Who would that freeze apply to – competitors only, perhaps? – and would outwardly ethical labs – Anthropic, perhaps? – be exempted?

The company has mastered the posture of proclaiming its own fright at just how powerful its technology is, implying by the strength of its own executives’ reactions that its model Claude is, in fact, mighty.

Anthropic’s branding as a lab populated by an army of coding Victor Frankensteins, so brilliant but so afraid of their own creations, has been successful. It poses as the more thoughtful counterweight to a reckless OpenAI, hellbent on releasing new products no matter their consequences, but competes on a similar product release schedule. The company’s slow rollout of its Mythos cybersecurity product only to trusted partners does signal a desire to responsibly roll AI out, but OpenAI followed the same playbook with its cybersecurity-focused product.

The question I keep coming back to, though, is this: with all this hand-wringing over AI safety, why keep developing AI? It’s not a compulsory service. Go do something else if you’re so afraid. But that seems unlikely, and you wouldn’t get rich from one of the stock market’s biggest debuts ever if you walked away now, and so I don’t think we need to accept the calls for a “pause” at face value.

The AI boom’s aftereffects