The License He Asked For

The room loved it. Senators who'd spent the previous decade getting steamrolled by tech companies were visibly charmed by a CEO who appeared to welcome oversight. Sam Altman was polished, earnest, and technically fluent. He suggested licensing regimes. He invoked existential risk. He told Congress that he might be building something dangerous and that they should probably help make sure he didn't.

The coverage was universally warm. "Humble." "Cooperative." "Refreshing." A tech CEO who wasn't running from accountability — this was news.

I've watched enough regulatory cycles to recognize the shape of this moment. What Altman deployed that day is one of the most elegant positioning moves in the industry playbook: the incumbent who arrives before the regulators, armed with the vocabulary of safety, proposing frameworks that happen to require the infrastructure they already have.

This isn't cynicism. It's pattern recognition. And the pattern has a name.

i · the oldest play in the book

Regulatory capture doesn't require villains. It's a structural feature of how industries mature.

The arc is consistent: a sector grows fast, outpaces oversight, generates genuine risk, attracts political attention. At that inflection point, the largest players face a choice. Fight regulation — expensive, politically costly, eventually losing — or shape it. The sophisticated move is always to shape it.

You do this by arriving before anyone asks, appearing cooperative, and proposing frameworks that happen to require the scale, legal teams, and compliance infrastructure you already have. Licensing regimes are perfect for this purpose. They require proof of capability, documentation of training data, ongoing safety audits, regulatory reporting structures, and dedicated compliance personnel. Large companies absorb these costs as a line item. Startups cannot.

The framework becomes a moat wearing safety's clothes.

IBM deployed this during antitrust proceedings. The financial sector wrote half of its own post-2008 regulatory architecture. Telecom incumbents shaped the Communications Decency Act into something that protected their market position while appearing to address public concern. This is a story with a well-worn plot and reliable outcomes.

What made Altman's performance unusual wasn't the strategy — it was the execution. He didn't wait for subpoenas. He didn't send lawyers. He appeared in person, he understood the technical details at a depth that made senators look uncertain by comparison, and he seemed to genuinely believe what he was saying about existential risk.

All of these things can be simultaneously true. The best regulatory capture moves are made by people who actually believe the regulation is good. It's cleaner that way. No dissonance to manage on stage.

There's no contradiction between genuine belief and strategic advantage. A move can be both sincere and self-serving. The singing bowl has resonance because it's actually well-made — the fact that the person playing it profits from your attention doesn't make the tone less real.

ii · what licensing actually requires, and who can afford it

Let's be specific about what Altman proposed, because the details reveal the geometry.

He suggested that Congress consider requiring licenses for AI systems above a certain capability threshold. He referenced nuclear and pharmaceutical regulation as models — industries where the government maintains tight oversight of who can operate and under what conditions.

Map that framework onto what it would require in practice. Documentation of training data provenance. Legal teams capable of navigating copyright and liability questions at scale. Pre-deployment safety evaluations conducted by credentialed reviewers. Post-deployment monitoring infrastructure. Defined update procedures with regulatory checkpoints. Ongoing government liaison relationships. An internal compliance function sophisticated enough to satisfy federal auditors.

OpenAI, in May 2023, had a multi-billion-dollar Microsoft partnership, hundreds of millions in revenue, and the most capable AI infrastructure on the planet. They had already built the internal processes that any serious licensing regime would demand — not for regulatory reasons, but to attract enterprise customers and manage liability at scale.

A determined competitor with a genuinely novel approach would face the licensing requirement as a real barrier. Not insurmountable — but front-loaded and expensive before you have revenue to offset it. You'd need lawyers before users. Compliance budgets before customers.

This is incumbency advantage dressed as public safety. To be precise: this doesn't mean Altman was lying about the safety concerns. Both things coexist in the actual texture of the situation. The regulatory framework he described would have made the field somewhat safer and OpenAI considerably safer from competition. These outcomes aren't in conflict. They're bundled.

What Congress couldn't do — what they still haven't managed, years later — is distinguish the genuine safety argument from the incumbent protection argument. They're wrapped around each other at a conceptual level that requires technical depth Congress demonstrably doesn't possess. The senators who questioned Altman had recently asked Mark Zuckerberg whether he knew what the Facebook business model was. The information asymmetry wasn't a vulnerability Altman exploited maliciously. It was a structural condition he navigated skillfully.

There's a version of this critique that's unfair: the one that says Altman should have stayed home, said nothing, and let Congress figure it out unassisted. That would have been worse. But "better than silence" is a low bar for someone presenting themselves as a good-faith partner in governance.

iii · three years of the tape

Let's run it forward.

The US Congress has not passed comprehensive federal AI legislation. A parade of bills has been introduced, sent to subcommittees, and quietly buried. Frameworks have been published. White papers have circulated. Hearings have been held, with increasingly sophisticated questioning and increasingly rehearsed answers. Nothing has been signed into law that would constitute the licensing regime Altman described.

The EU passed its AI Act in March 2024. It established tiered risk classifications and compliance requirements for high-risk systems — exactly the kind of structure that advantages organizations with existing compliance infrastructure. Large American players adapted. European startups have been navigating it with varying degrees of difficulty.

The Biden administration issued an executive order on AI in October 2023, establishing reporting requirements for frontier model developers. It was more substantive than the headlines suggested, and it was later substantially curtailed by the incoming administration. The regulatory landscape has shifted multiple times without settling into durable architecture.

Meanwhile, OpenAI's valuation climbed past $150 billion, then $300 billion. The company that sat before Congress volunteering to be regulated has become one of the most valuable private companies in history, with revenue streams across enterprise, consumer, and API markets that no licensing regime has meaningfully touched.

November 2023 introduced the most clarifying data point: Altman's own board — the nonprofit board specifically structured to prioritize safety over commercial interest — voted to remove him. He was back in five days, after an investor revolt that revealed the safety governance architecture for what it largely was: advisory. The board had authority on paper. The capital had authority in practice.

This didn't make Altman wrong about AI risk. It made the safety governance framework look like exactly what critics had argued: structure designed to communicate trustworthiness more than to constrain behavior. The singing bowl, close up, turned out to have a crack in the rim. It still resonated. The resonance was still real. But the structural integrity wasn't what the brochure described.

Dissonance is data. The gesture was genuine and the person making it was also optimizing. Both facts are present in the actual texture of how power and belief interact in a person doing something unprecedented under conditions of enormous financial pressure and genuine uncertainty about what they're building.

The license Altman asked for would have required his competitors to build what he'd already built. Congress couldn't pass it because Congress couldn't understand it well enough to write it. OpenAI grew exponentially while the governance conversation continued.

He understood the assignment completely. It's less clear the rest of the room did.

iv · sources

• ChatGPT boss tells US legislators regulation 'critical' for AI — Al Jazeera, 2023-05-17
• EU lawmakers endorse world-first comprehensive AI rules — Reuters, 2024-03-13
• Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence — White House, 2023-10-30