I’ve covered tech policy long enough to know that governments usually move at the speed of paperwork, not at the speed of code. But with artificial intelligence, the mood has flipped.
In the last two years I’ve watched lawmakers who can barely manage a microphone suddenly talk fluently about frontier models, red-teaming and systemic risk. Washington signs sweeping executive orders one month, Brussels rolls out a continent-wide rulebook the next, London hosts AI safety summits, and Beijing quietly publishes rules that can shutter a model overnight.
The message is clear: governments think AI can go off the rails badly enough to warrant emergency-style regulation. And for once, the political panic might be justified.
Europe’s AI Act: The World’s First Full-Stack Rulebook
If there’s a regulatory front-runner, it’s the European Union. Its AI Act—Regulation (EU) 2024/1689—is widely described as the world’s first comprehensive legal framework for AI. It doesn’t just nibble at the edges; it classifies systems by risk, bans certain uses outright and demands strict controls for “high-risk” applications like hiring tools, credit scoring and biometric surveillance.
Several provisions are already live, including bans on things like social scoring and some forms of real-time biometric tracking. The rest, including heavy compliance requirements for high-risk systems, will phase in through 2026.
From where I sit, the EU isn’t just regulating; it’s trying to export its values by regulation. If you sell AI into the EU, you effectively have to play by European rules. That’s why U.S. and Asian companies are quietly rewriting their internal policies to “AI Act-proof” their products.
The United States: One Executive Order, Two Administrations, and a Lot of Uncertainty
Across the Atlantic, the U.S. started with a big swing. In October 2023, President Biden signed Executive Order 14110 on the “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence,” instructing agencies to tackle everything from model safety tests to discrimination in AI-driven decisions.
The order pushed for:
- safety standards for powerful models,
- reporting requirements for big training runs, and
- protections around civil rights, worker displacement and national security.
Then politics happened. When the administration changed in January 2025, the new president rescinded the order on day one, arguing it “smothered innovation with red tape.”
From my vantage point, that didn’t kill AI governance in the U.S.—it just drove it sideways. Agencies are still drafting sector-specific rules, states are rolling out their own AI laws, and Congress is toying with bipartisan bills on deepfakes and critical infrastructure. But there’s no longer a single, unified federal blueprint.
In other words: Washington admits AI is dangerous, but it’s arguing over whether the bigger risk is runaway algorithms—or over-eager regulators.
The UK: From Bletchley Park to “Where Are the Laws?”
The UK, meanwhile, wants to brand itself as a global AI safety hub. In 2023 it hosted the first AI Safety Summit at Bletchley Park, bringing together 28 countries and major tech firms to sign the Bletchley Declaration—a statement that frontier AI could pose catastrophic risks and needs coordinated oversight.
Since then we’ve had a second summit in Seoul and a third announced for France, turning these meetings into a sort of “Davos for existential dread.”
Back home, though, the UK has been slower to translate summit talk into binding law. That’s changing. This month, more than 100 MPs and peers signed a letter urging the government to regulate the most powerful AI systems with controls on par with nuclear oversight, warning that advanced models could threaten national and global security if left unchecked.
When safety researchers, former defense officials and current lawmakers all start using phrases like “comparable to weapons of mass destruction”, you know the mood has shifted from academic debate to emergency planning.
China: Control First, Innovation Second
While Western capitals argue about rights and innovation, Beijing has taken a more straightforward route: regulate first, ask questions later.
In July 2023, China’s Cyberspace Administration and six other agencies released Interim Measures for the Management of Generative Artificial Intelligence Services, the country’s first dedicated rule set for public-facing generative AI.
The rules require providers to:
- register their models with authorities,
- conduct security assessments before launch, and
- ensure that generated content reflects “socialist core values” and does not undermine state security or social stability.
From my perspective, China’s approach treats AI both as a strategic industry and as a potential vector for political instability. It’s less about protecting individual rights and more about keeping information flows—and powerful models—under tight state control.
Why Everyone Is Suddenly Scared of the Same Things
Despite wildly different political systems, I keep hearing the same list of fears in policy papers and off-the-record interviews:
- Deepfakes and disinformation that can nuke trust in elections, markets and media overnight.
- Cyber and bio-security risks, where models help bad actors design malware or even biological agents.
- Runaway economic disruption, as AI eats into white-collar work faster than governments can retrain workers.
- Loss of human control over increasingly autonomous systems—what the UK summits politely describe as “frontier risk.”
Until recently, policymakers could treat AI like just another tech buzzword. Now they’re seeing concrete failures: chatbots that hallucinate libel, recommendation systems that amplify hate speech, and automated tools that quietly discriminate in hiring or lending.
The scary part? Most of this is happening with today’s systems, not the hypothetical “superintelligence” people argue about on podcasts.
The Global Rulebook Is Being Written—Whether You Like It or Not
From my seat at Abacus News, the story isn’t just “governments are panicking.” It’s that a de-facto global rulebook for AI is being drafted in real time, and if you build or deploy AI, you’re going to feel it.
A few practical realities jump out:
- If you sell into Europe, the EU AI Act’s risk categories and documentation requirements will likely become your baseline—even if you’re based in California or Shenzhen.
- If you operate in China, the content and security rules for generative AI aren’t optional; they’re existential.
- If you’re in the U.S. or UK, you’re entering a world of sector-specific rules, voluntary codes that are voluntary until they aren’t, and regulators that are suddenly very interested in your model cards and training data.
For developers and founders, this means AI governance can’t be an afterthought. “We’ll figure out compliance later” is starting to sound as reckless as “we’ll patch the brakes once the car is on the highway.”
The Real Risk Isn’t Too Many Rules, It’s the Wrong Ones
Do I worry about over-regulation? Honestly, yes. I’ve seen clumsy tech laws freeze innovation before. If governments treat every recommendation model like a nuclear reactor, they’ll suffocate the small players and entrench the giants.
But after reading the EU’s legal texts, the U.S. executive order, China’s interim measures and the Bletchley Declaration back-to-back, what strikes me most is not overreach—it’s how late we are to this conversation.
AI systems are already embedded in healthcare, hiring, finance, policing and national security. We didn’t just cross the Rubicon; we sprinted across while arguing about VPN bans. The regulatory scramble we’re seeing now is less “panic about hypothetical robots” and more a belated attempt to put guardrails on machinery that’s already running at full speed.
For readers of Abacus News—founders, policymakers, and everyday users—the takeaway is simple:
AI governance is no longer a niche legal issue. It’s the next big variable in how quickly, and safely, this technology remakes our lives.
If the rules are smart, they’ll force transparency, curb the worst abuses and keep the door open for innovation. If they’re clumsy, we’ll get the worst of both worlds: unsafe systems and stifled competition.
The alarm has been sounded. Now we get to see whether the people writing the rulebook understand the code they’re trying to control.
