'Government Has to Be Regulating This': Obama's Colgate Filibuster on AI Is the Sharpest Political Framing of the Debate Yet

Former President Barack Obama walked onto the stage at Colgate University’s Family Weekend conversation on Friday evening expecting to talk about leadership, public service, and the Obama Presidential Center. He got the AI question instead — and rather than deflect, he warned the moderator, “I’m a former senator. This might feel like a filibuster,” and then delivered roughly ten minutes that may be the most coherent political framing of the AI debate any national figure has offered to date.

The full transcript, published on his Medium account the next day, shows a politician who has clearly done the reading. And in a week when the frontier labs themselves called for a slowdown, Obama’s answer landed as both a policy argument and a not-so-subtle 2026 midterm message: “If somebody does not have a serious plan for how to deal with this, then they’re not meeting the moment, and you should probably look for somebody else.”

“The technology itself is not overhyped”

Obama opened by drawing a distinction that most political commentary on AI still gets wrong. The commercial benefits of AI may be overhyped, he said. The valuations of AI companies may be overhyped. “The technology itself is not overhyped.”

He reminded the audience that he had been tracking the issue since 2015–16, when his administration pulled together a commission on the political, social, and economic implications of AI and produced what he called “a terrific report” — one that, in his telling, ended up “in the same bin as my pandemic proposal,” ignored by the administration that followed.

The core technical claim of his answer concerned what he called recursive learning: the shift toward models that teach themselves. Using a teacher-student metaphor, Obama sketched the curve that has AI researchers spooked. As recently as a year ago, he estimated, roughly 90 percent of what frontier models learned was supplied by humans, with the models filling in perhaps 5 to 10 percent themselves. Today, he said, the ratio is around 50/50 — and if trend lines hold, “it might be 90/10,” with machines supplying the overwhelming majority of their own learning. That, he argued, is why the industry has hit the hockey-stick part of the trajectory, and why choices that could once be deferred now cannot be.

The threat model: not killer robots, but “bad humans”

Obama was careful to triage the risks. The science-fiction scenario — models smarter than humans that decide humans are “fine, but not necessary” — got a measured treatment: “I do not want to exaggerate that particular risk, but I will say there is a non-zero chance of that happening.” He noted that the danger there isn’t consciousness or malice but misalignment: systems setting their own agendas in ways that diverge from what humans actually want.

The risk he takes more seriously is older and simpler. “These models are getting powerful enough that if they get in the hands of bad humans, they can do bad things.” His example was blunt: somebody could take the equivalent of a future Claude or OpenAI model and ask it to design a new strain of smallpox releasable in the New York subway system “using ingredients I can buy at Home Depot.” A less apocalyptic variant — an agent instructed to maximize stock-market returns “and if the instructions are illegal, don’t worry about it” — could in his telling bring down the market.

Beyond catastrophe, he flagged economic disruption with a parable aimed squarely at his audience of soon-to-graduate families: the law-firm pyramid. Partners bill junior associates at steep rates; those associates endure the grind on the promise of reaching the top. AI investment in law firms, he observed, is rational precisely because it lets firms operate without the associates — and that is a “highly skilled, highly trained” job already in the blast radius. Add AI companions engineered to be as addictive as social media, aimed at children, and the psychological surface area of the problem expands further.

Why he believes the labs’ slowdown is sincere

The most politically interesting passage addressed skepticism on the left. Obama acknowledged the distrust of companies whose leaders suddenly discovered safety — noting pointedly that Anthropic and OpenAI are “about to launch massive IPOs that make all their employees millions and billions of dollars,” and that it is unprecedented for companies in that position to say their product may be dangerous.

His verdict: “In this situation, they are genuinely worried about their product.” Slowing down on the eve of an IPO, he argued, is about the least convenient position a company can take — and therefore a credible signal.

That leads to his short-term prescription. Some on the left, he said, argue government shouldn’t allow the companies to make that decision at all. He agreed that voluntary restraint “can’t be a long-term solution” — “Government has to be regulating this” — but with Congress inert and a White House he described as unwilling, voluntary industry restraint is “probably the best we can do for now.”

“That’s not how we treat airlines, or drug companies, or food companies”

The line that dominated the weekend’s clip economy came when Obama recounted hearing one of Donald Trump’s main AI advisers argue that the market will take care of safety: companies have every incentive to avoid danger, and if they fail, liability lawsuits will discipline them.

His rebuttal was a century and a half of regulatory history compressed into one sentence: “That’s not how we treat airlines, or drug companies, or food companies. We now have 150 years of experience just letting the market do whatever it wants when it comes to things that deeply affect our health and our safety.”

He also dismissed state-by-state regulation as structurally inadequate, comparing it to gun control in Illinois — strict laws at home mean little when neighboring states don’t follow, “and this is more dangerous and it moves faster.”

The second alignment problem

The most conceptually original part of the answer was what Obama called a second alignment problem — beyond the technical one of ensuring models obey intent. This one is the “misalignment between how this technology can be best used for society and the commercial imperatives of these companies.”

The AI buildout, he noted, has attracted more capital than the Apollo program — “one of the biggest, if not the biggest, investment in infrastructure and science and technology in human history.” Investors want returns, and that pressure forces companies to ship the thing that justifies the valuation: agents. Agentic AI — his references were JARVIS from Iron Man and Samantha from Her — is where the money is, because it replaces the junior associate and the accountant.

The problem, in his account, is that you test agents by “letting them run wild in the world,” unboxed, on the internet, looking at your bank accounts. And if the goal were simply curing cancer or solving zero-carbon energy, “you can do that without having agentic AI and having it just roaming free in the internet.” The reason agentic AI leads the roadmap is that it’s the product people will pay for — a structural incentive misfire, not necessarily villainy.

The political bet

Obama closed the filibuster with an appeal that doubles as electoral strategy: the country needs “a competent government and a serious bipartisan conversation around this issue, and we have to do it fast.” Voters, he said, should treat a serious AI plan as a threshold qualification for office.

Coming a week after his private fundraiser remarks urging Democrats to prioritize AI policy, and days after statements on the Obama Foundation site calling for public oversight, the Colgate answer confirms a deliberate escalation. Whether the “regulate it like airlines” frame survives contact with a Republican Party that has largely adopted accelerationist talking points — and an AI industry that split over the slowdown calls — will define one of the stranger policy fights of the 2026 midterms.