The AI Policy Landscape, Six Months On

Back in April, I wrote a piece for The Governance Post explaining the policy debates surrounding artificial intelligence I thought everyone should know about. Will AI take your job? What are data centres, and why do they matter? How are we regulating this technology? And what are the existential risks posed by the growing capabilities of artificial general intelligence (AGI)?

Working in tech policy, it has been interesting to see which debates have resurfaced and which have faded. Now seems like the right time to revisit my earlier claims and see which ones have stood the test of time.

AI and the Economy

In April, I argued that AI would probably not take your entire career, but it might take the entry-level job you were hoping to start with. Students have only become more worried since. A survey by Monster found that 89% of graduates now fear AI will replace entry-level roles, an increase from 64% in 2025. Employers are not helping either: in July, a survey of 1,000 US hiring managers showed that almost half would rather invest in AI tools than hire and train a graduate in the future.

However, the picture is messier than these numbers suggest. A September review of US hiring data showed an increased hiring of experienced workers, with AI being just one of several potential factors outlined in the review, including remote work and changes in the economy. The squeeze on graduates is real, but blaming it solely on AI feels overly reductive.

What concerns me more than the increasing body of literature on the future of work is the response of our educational institutions to these challenges. Only 29% of US graduates said their universities had given them extensive training in using AI at work. I have seen this myself at the Hertie School, where the only AI training on offer is a basic IT skills session. This is worrying for a policy school where many students hope to become researchers, given that AI is already reshaping the field. The onus falls on the individual, and those who cannot use the technology effectively risk being left behind.

Infrastructure and Regulation

My previous piece framed climate goals and AI infrastructure as being on a ‘collision course’. Data centres have since become one of the most heated local political fights in the United States. By one estimate, opponents blocked or delayed 75 projects worth $130 billion in the first three months of 2026 alone. The hostility is remarkably bipartisan: conservatives worry about property rights and electricity bills, environmental groups about water and emissions. A September 2026 poll by the University of Massachusetts Amherst found that just 11% of Americans backed building a data centre in their area.

In a moment that went viral in late May, Alexandria Ocasio-Cortez held up two jars of muddy water at a congressional hearing, the alleged drinking water left for residents of Morgan County, Georgia, following the construction of a large Meta datacentre. I expect this same sentiment to grow in Europe, as protests over the European Union’s own data centre buildout are already emerging. Just last year, the city of Amsterdam banned all new data centre development until 2030.

Previously, I presented the EU AI Act as the world’s first comprehensive AI law. My understanding has changed since, and much of how AI is actually governed in Europe runs through existing laws such as the Digital Services Act and the Digital Markets Act. These existing laws should be applied wherever the outcomes require it, without treating AI as a special case. However, how to govern the technology before models come to market is still a very live debate. In June, the EU signed off on the Digital Omnibus, which actually pushed back the rules for high-risk systems–such as AI used in hiring, credit scoring and education–to December 2027. Therefore, it is hard to depict the EU as championing AI regulation when it keeps moving its own deadlines.

In the US, the picture is just as contradictory. In June, the Department of Commerce used export controls to order Anthropic to cut off every foreign national from its two newest models, Mythos 5 and Fable 5, reportedly after a jailbreak raised national security concerns. The order was lifted on 30 June, but the lesson remains. The US government will step in when it sees a security threat, or simply when it wants to assert its grip over this technology.

AGI and Existential Risk

In my opinion, this is the policy debate which has changed the least since April. The same people are still sounding the alarm, and every few months someone working in AI safety quits and goes public with their fears of the end of the world. My own view has shifted somewhat. Whilst I still think existential risk should be taken seriously, for me the most pressing existential risk that AI poses is the concentration of vast power in the hands of a few. In this particular context, the ‘few’ in question have morally questionable track records, with some presenting as frankly unstable.

Reflections

So what did I get wrong? I feel I underestimated how willing the US is to intervene in the tech sector. Trump seems far less cosy with the tech billionaires than he was six months ago. I also gave the EU more credit for leadership than it has since earned. And whilst my claim that ‘no global framework exists’ is out of date on paper, it is still very much true in practice. Nothing binds countries, which is scary, if understandable, as the rules-based international order continues to crumble in front of our eyes.

When I titled my last piece ‘Bursting the AI Bubble’, the bubble I was referring to was the jargon that prevents most people from engaging with these policy debates. Today, when you put the words ‘AI’ and ‘bubble’ side by side, they mean something very different.

Put simply, AI as a product is not making money. The companies at the centre of the boom are spending far more than they earn. OpenAI was reported to be projecting losses of around $14 billion this year against roughly $20 billion in revenue, and it does not expect to turn a profit until 2029 at the earliest. Much of the money keeping the industry afloat also moves in a circle: Nvidia invests in OpenAI and OpenAI spends that money on Nvidia chips, while hyperscalers like Microsoft and Amazon invest in AI labs that then buy their cloud capacity. Critics describe this as circular financing and compare it to the dot-com era, whilst defenders say it is ordinary practice in the industry.

The mood in the tech sector seems reminiscent of that of the banking sector in 2007. As former chief executive of Citigroup Chuck Prince told the Financial Times in July 2007, ‘As long as the music is playing, you’ve got to get up and dance’. Prince resigned four months later and Citigroup went on to receive $45 billion of taxpayers’ money in a bailout. By March 2009, its share price had fallen below $1, wiping out more than $270 billion in value from ordinary people’s savings and pensions.

According to Goldman Sachs, AI-linked companies now make up nearly 45% of the S&P 500, the index behind millions of pensions and savings accounts, including in Europe. If the bubble bursts, then it will not only be investors in Nvidia and OpenAI who pay the price.

I am reminded of the ending of the film The Big Short. As the 2008 crash unfolds, the few investors who bet against the housing market make a fortune, but none of them feel like celebrating. Brad Pitt’s character reminds his young colleagues that every 1% increase in unemployment means tens of thousands of deaths. The narrator predicts that people will do what they always do and ‘blame immigrants and poor people’. If this bubble does burst, governments must have the power and the will to make sure those who inflated it pay the price, not ordinary people.

If any of the above interests you, get involved with the Hertie AI Policy Initiative. We run regular events with researchers and policymakers and are always looking for new writers, editors and event organisers.

Bruno Steel (MPP 2025) was born and raised in Edinburgh, Scotland, and currently works at Interface, a tech policy think tank in Berlin. He is also one of the Directors of the Hertie AI Policy Initiative. His policy interests span all things AI, from platform regulation to the impact of technology on society. These are informed by his experience in the humanitarian and third sectors. In his spare time he enjoys climbing mountains in the Scottish Highlands.