The 2028 presidential primaries could be crucial for AI outcomes
When I talk about my job and my concerns about AI killing everyone (after it takes a bunch of jobs), the more resilient people ask “what can we do about it?” My answer has become: “Elect a next president who shares these concerns, from either party.” I hope you’ll either join me in this prescription, or point out why I’m wrong. Below is the logic, in brief.
The US public really doesn’t like AI,[1]
even before clear job losses, more warning shots, and more expert concern as capabilities improve
AI caution is a likely platform of the next presidential campaign
If the next president is sincerely concerned and/or thoughtful
They could easily slow development and force safety measures
This would improve the odds of good AI outcomes
Anyone can be recruited/converted to spread awareness of AI risks, and otherwise help to elect a next president who’s genuinely aware of and concerned about AI risks. Strategy is out of scope for this brief post, but accurately and broadly conveying your and the field’s concerns seems like a useful push.
That’s a telegraphic form of the argument for spending some time and thought on the next US presidential primary. The remainder elaborates a little, by addressing some counterarguments and blockers.
Why not?
AI safety should remain non-partisan:
AI concerns among voters are currently mostly non-partisan[1]
Focus on conservative arguments/targets could maintain balance
Could increase likelihood of power concentration:
Valid concern but:
I think hoping the government just won’t notice AI potential is hopium
All the more reason to elect a trustworthy admin/government
Timelines could be earlier/later:
Possible. The most careful predictions (e.g., AI 2040) put much of the probability mass in the 2029-2033 presidential term
The public is already concerned:
Yes, but more and better-focused is better:
“Yes, job loss could be severe, AND average expert and developer opinion is that it might well kill us all”
Public statements don’t imply useful action:
Exactly! Candidates will adopt AI-caution rhetoric by default in the general election, so the real choice happens in the primaries (both open in 2028).
No actually-concerned candidates to support:
Exactly! Now through 2027 is the time to find or convince candidates
Slowing down means losing to China:
Racing flat-out isn’t inevitable; a negotiated mutual slowdown seems quite plausible.[2]
The next administration will treat AI like this one does:
Not inevitable, IMO not even likely, due to some combination of
Public outcry over job losses
Software security concerns, as with the Mythos release
Loss-of-control fears in the national security apparatus, the public, the president, or their close advisors
Below I’ll say a little more about some of these blockers; if you’re convinced, you could skip this.
The president won’t take action
It’s easy to assume that the government is unlikely to exert control over AGI projects in time to matter. I don’t think that holds up. To me the current situation looks much like expressing disbelief and despair that the government won’t do anything about COVID in January 2020. Public opinion changes nonlinearly, but it does change, and if anything it tilts toward panic. People, including politicians, don’t often extrapolate trends, but they do respond when dangers become obvious. We haven’t even seen discernible job loss yet, and public opinion already weighs far toward concern and caution.
I’d go further: it seems not just likely but probable that the government will assert control over a crucial matter of collective security. This is its intended role and it has done so historically. We have a large and competent section of the government whose job is looking for imminent dangers and making sure the President authorizes appropriate actions (the national security apparatus), and large portions of that are alert to subtle, hidden, and long-term threats.
I take the inaction and disinterest to date as little evidence that the government won’t intervene in time. There has been near zero danger from AI so far. When that seemed to change, claims of Mythos’ security capabilities were the first sign of looming trouble, and the government acted immediately and strongly, if capriciously.
The president can’t take action
One inspiration for writing this short post was the recent, excellent post How the American Executive Could Control AI Companies. This expanded on the arguments I briefly made in Whether governments will control AGI is important and neglected. In short: there are a lot of ways, even without congressional authorization or extralegal maneuvers. The executive has historically exerted control in matters of urgency, since the other branches of government are much slower to act. Even limited slowdowns, coordinations, or temporary bans on job-replacing AI[3] could blunt the worst risks from racing.
The public doesn’t care enough to elect an AI-concerned president
The public doesn’t care enough now, but they haven’t yet seen real job losses or real hacking incidents that affect them. Last-minute is the human way; we are busy and we discount uncertain futures heavily. In A country of alien idiots in a datacenter: AI progress and public alarm I lay out a scenario in which misbehavior from agents plays a central role.
I don’t think this is a long shot. I think it’s near a tipping point and could be relatively easily pushed. Even now the public is far more concerned than optimistic about AI. If there are more incidents like the Hugging Face hack or the government intervention on the Fable release, concern about AI will likely be a necessary part of a successful platform. And even if you find a strong change in public attitude unlikely, it seems worth playing the odds and putting some effort into spreading awareness, and finding and nominating a good candidate.
Thanks to Viktor Bowallius, Peter Gebauer, and Richard Juggins for insightful comments on earlier drafts.
- ^
The AI Safety Majority by the AI Policy Institute, June 2026, is the best poll I’ve seen to date. It shows concern outstripping excitement about AI by 3:1, and mandatory safety standards beating no regulation by roughly 10:1. See the section “Partisan differences are negligible.”
- ^
Xi Jinping’s statements suggest understanding and real concern over loss of control, both to AI itself and to an AI-empowered populace. For instance, his keynote at the 2026 World AI Conference calls for measures to “forestall loss of control” and to “ensure that AI is always under human control”. Even coordination much simpler than that in AI 2040: Plan A, like government-mandated safety measures on both sides and lab-to-lab lines of communication on alignment, would reduce the worst pressures and risks from racing.
- ^
Public outcry over job losses could lead to a temporary ban on release of systems that can take full jobs, and this could slow progress to AGI. IMO improving continual learning and metacognition is the most likely path both to job loss and to AGI; see LLM AGI will have memory and Human-like metacognitive skills will reduce LLM slop.
I’ve said before that we generally overestimate the salience of AI in American politics. It is widely-disliked, but it is also very low salience. Polls confirm this. We see:
A third of voters consider the economy most important, broken down into taxes, affordability, and the stock market.
Behind this are the usual slate of big issues that significant numbers of people care about. Immigration is the top issue for 7 percent, four percent are concerned about foreign policy, nine percent have some general issue with how the government is run. Four percent are sufficiently obsessed with Trump in one direction or another that they consider him, the person, as their foremost issue.
AI isn’t on the scoreboard. Lower down in the article, 22 percent agree with the Democrats, 23 percent agree with the Republicans, and 51 percent either don’t care or don’t see a difference.
Compare this to issues like immigration and crime, where Republicans hold a substantial advantage, and issues like environment and abortion, where Democrats hold a significant advantage. On all four, the ‘neither’ category is around 20 percent.
You cite job losses and hacking incidents as sources of future concern, but I think that the former will be outweighed by debates on immigration and things like the H1-B visa (barring a very significant spike in the rate of LLM improvement), and things like the latter are usually relatively low salience to ordinary people because of how few people they directly affect.
Put plainly, I do not think that AI decides a substantial number of primary votes.
The GOP primary will be between JD Vance, who will carefully toe the line between economic populists and technocrats with the aim of not making anyone too upset, and whichever unfortunate that the establishment coronates to run against him, who will quite possibly write a song about how much he loves data centers, labor visas, and a large military presence in the Middle East, but will lose handily even if he doesn’t.
The DNC primary could be one of two things.
It could be a 2016-style brawl between the left-populist wing, which will doubtless talk about AI as one among many niche issues, but will be more concerned with foreign war and general affordability, and the establishment, whose campaign will focus on dismissing the populists and insisting that they’re the only ones who can beat Super-Trump-Part-3. Generally, the establishment tends to win these things.
It could also be a 2020/2024-style coronation of a candidate that the establishment coordinates around. I consider this somewhat more probable, on the basis that, after 2016, the Bernie wing is somewhat fragmented and demoralized, and the single issue anti-Trump faction is pretty tightly managed, and likely to vote as a bloc.
More broadly, anyone that wants to make AI more salient could certainly make gains on the Right. Issues like this and this haven’t really gotten any press, for example. Not even a promise to Do Better from Anthropic et al.. On the Left, the establishment seems fairly satisfied with the political direction of most frontier companies, and the populists seem more concerned with AI, but for reasons outside the scope of the ones LW cares about.
If the US President comes to see AI as a potent danger to his family’s continued survival, then he will tend to act against it very vigorously even if most voters don’t have any strong feelings on the subject and even if the labs spend billions on lobbying. It’s not like most policy issues where those affected by the issue have to hope that policymakers who are insulated from the issue will empathize with them.
I’m hoping that the next President will be wicked smart or at least will surround himself with wicked smart people.
I do not think that you are modeling U.S. politics accurately. We are one of innumerably many niche interest groups who believe that their niche issue is the be-all-end-all most important thing in American politics, which presents the ultimate threat to life, liberty, and the pursuit of happiness, and that just showing the President our obviously true ideas will cause him to act.
Maybe we think we’re the only ones who are right, but that doesn’t change the fact that there are many such groups and, inevitably, any American politician with any degree of success has already met lots of them before, and, inevitably, any American politician who hasn’t burned out has learned to tune them out to a certain degree, by instinct.
Any policy objective should be motivated by practical concerns. Public examples of real people being harmed at scale by poorly-aligned LLMs, and the potential for incidents in the immediate future based on past incidents.
The hiring discrimination thing is a good example. Show some lawyer-ish GOP politicians concerned about illegal AA practices, point out that LLMs can be used by bad actors to launder banned discrimination, and there’s useful inroads to a check on AI expansion/deployment and potentially regulations on enterprise training/fine-tuning that involves hunting evidence of misalignment.
Similarly, you could show politicians the lives tradeoff thing and connect that to military deployments. Politicians presumably wouldn’t be thrilled about an AI tactical advisor that’s willing to get their enlisted constituents killed to preserve the lives of an upweighted demographic. If not, the right social media campaign, targeted to those same constituents, could make them care.
I should note that there is a wrong way to do this. Many interest groups, most recently the “Equity” people, are infamous for insisting that the solution to every problem from deficits to military readiness is solvable by funding their pet issue, and this has made them extraordinarily unpopular among people who see this is dishonest and disingenuous. A focus on immediate issues that actually have happened is vital, particularly given that fantastical claims[1] have already been a substantial impediment to safety campaigns.
False or otherwise; non-AI-researchers don’t have any means of distinguishing fantastical-but-true from fantastical-and-false, and countless grant-seeking researchers have played up the apocalyptic consequences of defunding them in the past.
I don’t consider any of the issues you mention to shed any light because none of them were ever potent threats the survival of the power elite. A nuclear exchange between the great powers does shed light (because while the power elite in both the US and the USSR would’ve probably survived a nuclear exchange initiated by their country, it would have severely impacted the quality of life of themselves and their families for many years) and of course no decisionmaker has ever chosen to initiate such a nuclear exchange.
If that is what you believe then I can see how you would end up believing the rest of what you wrote.
I’ve never been an AI researcher and I can correctly distinguish AI extinction danger as an actual potent extinction danger among the many candidates I’ve heard over my lifetime.
Holly Elmore (head of Pause AI US) has never been a researcher or engineer in AI or even in computing, and yet she’s been loudly proclaiming since 2023 that AI is a potent extinction danger whereas I’ve never heard her talk publicly about any extiction danger other than AI.
Jaan Tallinn has never been a researcher or engineer in AI (his technical career was in game development and peer-to-peer networking; he has a degree in physics) and yet he considers AI to be such a potent danger that he has spent $150 million of his money on AI safety. His spending on AI safety started early 2009 and continues.
Philosopher David Chalmers has never been anything like an AI researcher or engineer and yet he signed a public statement that “mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” So did many other non-AI-researchers.
That’s not a counterexample unless you can concretely claim that, had you stumbled onto any similar cause instead of AI Safety, you wouldn’t have taken up their crusade instead.
“Only experts can tell whether this is true” != “Everyone who believes this is true must be an expert”
Thanks for the detailed engagement!
Yeah. I do realize that AI isn’t currently very salient to many people, but the post doesn’t make that clear. Maybe I should’ve framed this more conditionally: if AI progresses, the issue could become salient enough to be a real factor, and we should prepare for that possibility. I think it’s fairly likely, particularly if timelines are short, but it’s just a possibility.
If the nominations turn out to be largely elite coronations, spreading the word on concern still works, the strategies just change somewhat. It will be harder to track sincerity, but getting elites concerned will still tend to trickle up.
AI probably won’t hit the top one or two concerns of either public or elites by the primaries. But it could. The spread of top concerns you list is pretty fragmented, so it wouldn’t take much for AI to reach immigration-level numbers, and that’s enough to change what candidates say and who they hire. Those in turn change their beliefs. I didn’t tackle this in the post, but I think that motivated reasoning tends to pull true beliefs toward stated beliefs in many cases. If you have no real opinion, saying you’re concerned will tend to make you concerned. And talking to hires who are truly concerned will definitely push in that direction.
WRT current vs. future salience: I feel like there’s an ambient opinion that the public won’t care about AI until it’s too late. But that’s not modeling the real possibility of rapid and nonlinear public opinion shifts, like the turnover in opinion and salience just before COVID lockdowns.Edit: the first half of today’s Warning Shots: A Theory makes this point succinctly. I was planning to write my own version as “minds change” but that expresses pretty much my exact argument.
Loss of control to AI isn’t hard to understand; it just seems like scifi, not immediate and so not worth worrying about, until you’re talking to a system that’s smart and agentic like you are. I think most people will have that experience to within one degree of separation by the election: a friend or family member telling them that AI is not just a dumb tool, and that it will take jobs if not take over. I just hope it happens in time for the primaries.
It’s strange to find myself hoping for faster timelines in that regard. But it might be worth it to get a next president who’s at least somewhat AI-concerned.
I replied to the other fellow with a practical strategy that doesn’t depend on anything shocking happening in the meantime:
Cool! Run me down the list of remotely viable presidential candidates who audibly care about AI x-risk and are unlikely to accelerate an arms race?
Jon Ossoff (currently at 11% odds to win the 2028 presidential election, per Kalshi) has pressed the director of the OSTP to implement mandatory DNA-synthesis screening in order to mitigate the threat of AI-enabled bioweapons development. And way back in 2023, just a few months after the release of GPT-4, he said that future AI could lead to “catastrophic political destabilization”, “cybersecurity threats”, etc, but might also “diagnose cancers” and produce “lifesaving drugs”. Seems like he’s taking AI somewhat seriously, as far as politicians go!
Gavin Newsom, currently at 9% odds to become president, vetoed the 2024 AI safety bill “SB 1047”, but signed the next year’s watered-down iteration, SB 53. It became the first law directly regulating frontier AI labs like Anthropic and OpenAI; apparently it “applies to models trained above 10^26 FLOP, with heavier duties on developers above $500M revenue; requires published safety frameworks and pre-deployment transparency reports; defines catastrophic risk as 50+ deaths or $1B+ damage, expert-level CBRN uplift, autonomous crime/cyberattack, or loss of developer control; requires critical-incident reporting to Cal OES within 15 days (24 hours if imminent); protects whistleblowers; caps penalties at $1M per violation; and creates CalCompute, a public compute consortium.”
AOC (13% chance to become president, highest of anyone except JD Vance) is in a weird situation where she has a huge datacenter-moratorium bill that would be a very extreme attack on the AI industry, almost trying to force a MIRI-style global pause on AI development. (An absolute ban on constructing or upgrading any datacenters in the USA—plus a ban on exporting GPUs to ANY nation, not just China—the datacenter moratorium would only be lifted once America has met a long checklist of passing laws to protect workers from being fired, an FDA-style system for pre-market review before labs are allowed to release models, etc. And the worldwide export ban would only be lifted once other nations had met the same policy checklist.)
And yet, unlike Bernie Sanders who talks all the time these days about AI as an existential risk, a massive-job-automation risk, etc, AOC herself is very dismissive of the actual technology of AI… she says it’s a bubble, worries about deepfakes, complains that palantir’s AI is helping ICE deport immigrants, etc. So she weirdly sees AI as a total nothingburger (just a subset of “big tech” issues or one aspect of “billionaire power” / “corporate power” generally, not a significant or consequential technology in its own right), even while simultaneously proposing the most extreme policy action on AI/datacenters as one of the centerpieces of her platform.
- All other dems have less than 5% odds to win. Mark Kelley (3% odds) is one member of this crowd, but has an interesting “AI for America roadmap”. It doesn’t treat AI as an existential risk at all, but has some interesting ideas about how to create a huge “AI Horizon Fund”, funded by AI industry revenues in various ways, that would be dedicated to mitigating job displacement and investing in infrastructure and so forth—the kind of thing feels like it could eventually transform into a “universal basic capital” scheme in some sort of AI-2040 takeoff scenario.
Meanwhile, the republican side is simpler:
- Vance (22% odds to be president) is quite engaged with AI (he’s read AI 2027!) and believes it will be a very big deal, but is a hardcore tech-accelerationist who has tended to dismiss safety concerns.
- Rubio (13%) has made almost no comments about AI whatsoever so it’s hard to tell what he thinks. So who knows what opinions he might come to after he starts thinking seriously about it. The one thing I’d say about him is that he’s a huge anti-communist China hawk with a long record of advocating human-rights-related sanctions against Xinjiang, support for Taiwan, etc. So regardless of how seriously he takes AI x-risk, he’d probably be very eager to support a Situational-Awareness-style / MAIM-style “entente strategy” of chip export controls and related pressure aimed at slowing China’s AI development.
- All other republicans have less than 5% odds to win.
So, on your three criteria of 1. remotely viable, 2. care about x-risk, 3. unlikely to accelerate an arms race:
- Jon Ossoff seems to care about AI x-risk but it’s unclear what his actual policy goals are aside from bioweapons and building-but-taxing datacenters, so idk about the arms race criterion.
- AOC would definitely try hard to slow down the pace of AI development (thus unlikely to accelerate an arms race), but she aggressively refuses to think about AI as an x-risk even while some allies, like Bernie Sanders, do.
- Your call whether Gavin Newsom sincerely cares about AI x-risk, or just signed SB53 without much thought because it struck him as a reasonable, low-downside compromise between safety activists and the AI industry. But SB53 strikes me as a good law. And he’s probably at least HEARD more AI x-risk discussion than most other candidates, since he’s been involved in numerous debates around various proposed AI safety bills.
Re: Seth Herd’s response “this isn’t about the candidates we have, but the candidates we can get”. I disagree with the implication that maybe we could find a great AGI-pilled politician and successfully elevate them to the presidency in only two years. (Alex Bores 2028, anyone??) The various prediction markets seem pretty confident that there’s an 80%+ chance that the next president will be drawn from a small group of 10-20 top contenders (people like JD Vance, Marco Rubio, AOC, Gavin Newsom, etc), and a less than 20% chance that some relatively-unknown dark-horse candidate will overtake them.
But I agree with Seth’s response in this alternative sense: two years is a lot of time for the current front-runners to learn a lot about AI issues, to change their positions and develop new policies, et cetera. AI will almost certainly be a vastly more important topic in 2028 than in 2026. So on both the democrat side and the republican side, there’s potentially a lot of room to educate and influence these leading contenders, shape their proposals, etc.
Great breakdown, thanks! You’ve done more homework than I have.
Yes, I agree that recruiting someone already AI-pilled is not a good strategy. I was thinking more of the strategy of pitching the arguments to viable candidates (along with pitching how those arguments could help them win). I was thinking more of that but mixing the two in my phrasing, as you noted.
I don’t know Ossoff or Newsom at all. AOC might actually be good by aggressively regulating AI for job loss purposes, which as I mentioend in a footnote would probably also slow progress to takeover-capable AGI since the two are roughly on the same path.
Vance is interesting. He’s very pro-AI and anti-regulation right now but he is someone who actually thinks and has publicly changed his mind on several important issues. And Fable thinks he’s probably not a sociopath or sadist, which is my single biggest criteria. (A sociopathic sadist in charge of the whole future would be the likeliest way to get a worse future than extinction IMO).
This isn’t about the candidates we have, but the candidates we can get if AI continues to progress, and we work for it.
I’m not very familiar with Republican or moderate Democrat candidates, but at least on the progressive Democrat side of things, Abdul El-Sayed has been surprisingly vocal about loss of control risks in his platform.
I would argue that a presidential candidate mentioning AI Safety in a speech would be extremely bad for the cause of AI Safety. Because if s/he mentions it in a speech, the goal will be to score points against the Enemy Team, and thus will tend to partisanize the issue.
Separately, I encourage AI Safety forecasters to dedicate copious amounts of brainpower to gaming out different strategies for AI regulation, and then present the best strategy to smart DC insiders, regardless of party, to implement behind the scenes. Many regulatory schemes will be calamitous, either backfiring or surrendering the lead in AI development to foreign rivals. But it’s plausible that smart rationalists could come up with a good plan.
Regulation is really a different topic from what I’m suggesting here.
I’m thinking of direct power from the president through executive orders and the executive branch’s various options, listed very nicely in the linked post and with the salient example of declaring Fable a supply chain risk. These are ways the executive can bully the industry. I think this could easily extend to a de-facto government takeover. Imagine companies producing some new superweapone. The government asserting control of how they do business, their security, and who they sell to would be seen as right and natural. It would be asserted initially by the executive branch and then ratified by congress if they even bothered. That is the actual situation with AGI projects. I think we should expect the executive and government at large to see this and treat it this way. Although I’m not sure this happens before we create TCAI and it does whatever it wants.
I have much less hope that regulations as such will be effective. They’re just too slow and clumsy. Unless they’re as general as “don’t create AI that can take full jobs” or “don’t create AI more generally capable than humans” whose enforcement would amount to judgment calls anyway.
This is a good point about competition and polarization as the politician’s goal. It’s less true of the primaries, where they’re mostly trying to defeat their party-mates, but still holds weight there.
It won’t be easy to polarize AI safety from here since dislike/distrust of AI is so remarkably evenly split across parties. but it could easily happen if one party (more likely the dems) picks up the banner and runs with it first. That’s why I suggested advocacy should focus on conservatives to maintain that balance.
Is there anything like a think tank that monitors policies and political parties, rating them on their commitment to or opposition to AI Safety? In France, something like this exists for animal welfare, and it’s insightful whilst sticking to a fact-based approach.
I don’t think there is. Doing this, even in an informal way, and having discussions about it here or elesewhere seems like a very useful thing to do.
https://www.datasecretslox.com/index.php/topic,16467.0.html
My take on your post is here, should you wish to read it, as I have a policy of telling people if I write about them.
That doesn’t seem like a very useful format for replies, and it didn’t seem like you’d put much thought into your replies there. Posting your most original and well thought out additions here would be a lot more useful.
So I’m not downvoting your comment but I’m not up orong it either.
Their critique was poorly presented, but I think I agree with a big part of it. Focusing on electing a good leader two years from now is strictly worse than focusing on political action now (e.g. resisting datacenters, resisting surveillance, lobbying for algorithmic liability).
Resisting surveillance seems like a good move. Resisting data centers doesnt; how is putting them elsewhere going to help?
Algorithmic liability seems like it will slow public deployment but also prevent a lot of warning shots, which are one of my two my main source of hope for surviving misalignment and takeover. But it would slow development a little, which would help a little.
I want politicians to lose popularity points supporting datacenters anywhere, and it has to start somewhere.
This argument seems backwards. Here’s an analogy: “If I drive without a seatbelt, then mild crashes will hurt me more, and thus do more to teach me to drive safely. If I wear a seatbelt, these valuable warning shots won’t happen, so my safety will be reduced.” But actually, if you drive with a seatbelt, your safety will be increased. The same is true for algorithmic liability: it will prevent some warning shots, but the increase in safety will more than make up for it.
I think it depends on whether you think alignment is on-track to succeed, or whether we need slodowns or new alignment approaches.
Algorithmic liability would increase focus on existing alignment techniques.( It would also, I think, lock away more frontier models from public use but we can set that aside). If those are adequate for future AI or will naturally lead to techniques that are, great! If they’re not, preventing misalignments now will prevent warning shots that could convince the public to support slowdowns and the alignment community to support new techniques that will actually work.
I’m not sure which I actually think. And I think nobody should be sure.
Which may be why the alignment community hasn’t lobbied more for algorithmic liability. Although it might have to do more with a generally libertarian philosophy.