Brendan Steinhauser
CEO, the Alliance for Secure AI
Place in the cosmos
Steinhauser is a conservative operative who hails from Texas GOP circles and still lives in Austin, but the alliance he runs is a bipartisan group deeply engaged in Washington.On “AI Safety”
“AI safety means ensuring that advanced AI is safe and secure, and allows human beings to reap the benefits of AI while mitigating the risks.”
In the battle over AI regulation — from Washington, D.C. to Silicon Valley — one term has become unexpectedly controversial: “AI safety.” After all, who doesn’t believe in making these tools safe?
But depending on where people stand on the issue, “AI safety” has come to signal wildly different things. Some are concerned about existential threats to humanity, while others are worried about protecting children and warding off mass job loss — and then there are those who think preaching “safety” is often a nonsensical excuse to shut down any technological progress at all.
“AI safety” is now something of a Rorschach test for how tech founders, politicians and everyone in between thinks about AI development. Understanding how that term has come to have different connotations for different stakeholders is a useful way to map the vast universe of AI policy and politics that has quickly emerged.
What’s perhaps most surprising is that the debate doesn’t break down along obvious partisan or ideological divides. The technology is still so young that political opinions surrounding AI have yet to be firmly locked in, and it’s possible to find skeptics and boosters on both the left and right.
Traditionally, the “safety-focused” community has been broadly associated with the effective altruist movement, which has significant ties to much of the AI world, in particular to the leading firm, Anthropic. But even among effective altruists, there is significant debate about how best to regulate AI — and as the issue has spread beyond Silicon Valley, where many of them reside, the concept of “AI safety” has grown more amorphous.
So, we decided to ask 20 of the smartest, most consequential people working in and around the industry the same question: When you hear the phrase “AI safety,” what’s your first reaction and what does it mean to you?
Here’s what they said, and where they fall into specialized categories created by POLITICO Magazine.
Doomsday Prophets
A group of mostly Bay Area-based AI researchers who believe the products their fellow technologists are building could imminently kill us all.
Nate Soares
President, Machine Intelligence Research Institute
Place in the cosmos
Soares, who co-wrote the book If Anyone Builds It, Everyone Dies, is among the tech universe’s most hardcore proponents of completely shutting down AI development.On “AI Safety”
“I’ve never liked the term ‘AI safety.’ If you’re in a car careening towards the edge of a cliff, it’s not the time to bring up ‘car safety.’ Car safety is about seat belts; it’s about airbags; it’s about crumple zones; and none of those will save you if the car goes soaring off the cliff. AI companies are racing to create artificial superintelligence, and the AIs they’re creating aren’t carefully programmed to follow instructions; they’re a mess of trained numbers that even the creators don’t understand. What we need is not a little extra ‘AI safety.’ What we need is to stop this reckless race.”
Max Tegmark
Physics professor, Massachusetts Institute of Technology, and co-founder, Future of Life Institute
Place in the cosmos
Tegmark is a Soares ally who has called his book the “most important of the decade.” The Future of Life Institute has also embraced effective altruism but has more revenue than MIRI, and in turn funds more projects that are similarly focused on catastrophic risk.On “AI Safety”
“My reaction is that it’s the latest phrase to be neutered by corporate lobbying. After greenwashing and ethics-washing came safety-washing from the top AI companies: They use the word ‘safety’ while racing to create AI girlfriends for children, deepfakes for fraud and uncontrollable superintelligence while lobbying against any meaningful accountability.”
Holly Elmore
Executive director, PauseAI U.S.
Place in the cosmos
Elmore is a proponent of merely pausing all AI development rather than stopping it altogether like Soares. She is also more directly engaged in popular protest around AI development, leading actions in both Silicon Valley and Washington.On “AI Safety”
“The thing that I’m advocating for is pausing frontier development and starting from there. Don’t make the problem any worse. We have as much time as we need at that point to figure out if we can make it safe. [Anthropic] used to call themselves an ‘AI safety’ company. Like I’d meet somebody at a party, and they’d say ‘I work at an “AI safety” company.’ It turned out, they meant Anthropic, which is laughable. That’s a frontier AI company.”
Scared Straight Caucus
People who are deeply fearful of the existential risks that AI poses, but who stop short of arguing human extinction is inevitable if industry keeps building.
Nick Bostrom
Philosopher on existential risk
Place in the cosmos
Bostrom is not directly involved with AI advocacy efforts, but writes and makes public appearances on the issue of existential risk. His 2014 book Superintelligence: Paths, Dangers, Strategies was a best seller and is widely read among Silicon Valley founders.On “AI Safety”
“I think primarily of the technical challenge of aligning superintelligence. That is: developing scalable methods for making AIs want to behave as intended that continue to work, even as their capabilities eventually exceed those of any human (including strategic thinking, persuasion ability, scientific creativity). There are other important challenges that we must also meet to achieve a positive outcome as we enter the machine intelligence era — various political and governance problems, and ethical problems such as how to make sure the future is also good for the digital minds that we create — but I place those under their own separate rubrics.”
Connor Leahy
U.S. Director, ControlAI
Place in the cosmos
ControlAI is a nonprofit primarily focused on warning politicians and the public about the risks of AI. The group built a coalition of over 110 British lawmakers who recognize superintelligence as a national security threat, and under Leahy, is now working to expand its footprint in the United States.On “AI Safety”
“Right now, ‘AI safety’ is a misnomer. Without international restrictions on developing superintelligence, an unchecked industry building autonomous AI risks human extinction — within a few short years.”
Safety-Conscious Right
Conservatives who fear the unchecked development of AI, both for existential reasons and for the ways it’s already affecting society.
Mark Beall
President, AI Policy Network
Place in the cosmos
Beall, a conservative political operative, runs the AI Policy Network, an influential, bipartisan group advocating for federal safeguards around AI. He represents a growing faction of conservatives who are increasingly concerned about the technology — and are fighting for influence within the Trump administration.On “AI Safety”
“‘AI safety’ means protecting America’s national security from advanced AI threats such as weaponization and loss of control. Current AIs, like Claude Mythos, already can hack critical infrastructure. Future AIs will likely be superintelligent, greatly exceeding human expertise. As we accelerate innovation and promote AI’s upside, ‘AI safety’ means having the technical capacity to ensure such systems remain under human stewardship.”
Evan Swarztrauber
Senior fellow, Digital Progress Institute
Place in the cosmos
Swarztrauber, who used to work for Trump FCC chiefs Brendan Carr and Ajit Pai, is also one of the leading conservative voices pushing AI guardrails. He is animated by issues like child safety more than existential risk, and he thinks that the idea of ‘AI safety’ has become politicized already.On “AI Safety”
“Because of politics, my initial reaction to ‘AI safety’ is often colored by lobbying groups that all define the term differently. But here’s what it should mean: rigorous safety testing for frontier models, robust protections for kids like chatbot age verification, and strong export controls to prevent hostile adversaries from acquiring dangerous capabilities with American technology. None of that should be controversial, even if the term ‘AI safety’ remains a hot button.”
Brendan Steinhauser
CEO, the Alliance for Secure AI
Place in the cosmos
Steinhauser is a conservative operative who hails from Texas GOP circles and still lives in Austin, but the alliance he runs is a bipartisan group deeply engaged in Washington.On “AI Safety”
“AI safety means ensuring that advanced AI is safe and secure, and allows human beings to reap the benefits of AI while mitigating the risks.”
Money Movers
Those within the broad “safety” community who are more mainstream than some AI skeptics — and have the cash to spend to support their pet causes.
Michele L. Jawando
CEO, Omidyar Network
Place in the cosmos
Jawando runs the Omidyar Network, a powerful “philanthropic investment firm” that funds nonprofits and for-profit companies around the globe. It is often praised by effective altruists and does work in San Francisco, but Jawando focuses largely on policymaking in Washington.On “AI Safety”
“AI safety is often framed as only a technical challenge, but at its core it’s a people challenge. It means building systems that don’t concentrate power or spread harm, but expand opportunity. It means technology that is trustworthy, transparent and accountable to the communities it affects.”
Brad Carson
President, Americans for Responsible Innovation
Place in the cosmos
A former Democratic congressman from Oklahoma, Carson is the co-founder of Public First Action, a super PAC network and advocacy group that has collected significant donations from Anthropic and is in the midst of an extended proxy war with Leading the Future, a rival super PAC network funded in large part by executives at the venture firm a16z and the AI company OpenAI.On “AI Safety”
“AI safety is an essential issue for both the citizenry and for the future of the AI industry itself. And it’s an issue that is going to redefine American politics in the 21st century.”
Politicians
People either in Congress or who have run for office who have been pressing for AI safety.
Bernie Sanders
Independent senator, Vermont
Place in the cosmos
Sanders has been one of the more outspoken lawmakers on the issue of AI, proposing a moratorium on data center construction and a dividend to help Americans secure financial benefits from AI companies.On “AI Safety”
“In America today, there are more health and safety regulations around opening up a new sandwich shop than releasing a new AI model that could literally transform what it means to be human. That’s insane. AI and robotics will have a profound impact on the life of every man, woman and child in our country. These revolutionary technologies could displace tens of millions of workers, cause enormous harm to our children’s emotional well-being and cognitive capabilities, destroy privacy as we know it, devastate the environment and even wipe out the human race. We need to make certain that AI works for all of us, not just the billionaires who own it.”
Jay Obernolte
Republican representative, California
Place in the cosmos
Obernolte, a former computer programmer and video game executive, recently released bipartisan legislation with Rep. Lori Trahan (D-Mass.) aimed at creating a federal framework for legislation around AI.On “AI Safety”
“When the House AI Task Force looked at the overall issue of AI safety through the 25 hearings that we conducted in the last Congress, the most pressing concern about safety involved the malicious use of AI by humans. So AI is a very powerful tool for enhancing human productivity, and as we say, ‘bad people are going to bad’ and they’re going to bad more productively with AI. So we have to be very concerned about putting in place guardrails that mitigate the use of AI by malicious actors.”
Andrew Yang
Entrepreneur, former candidate for president in the Democratic primary
Place in the cosmos
As a political newcomer who proposed ideas like a universal basic income during the 2020 Democratic primaries, Yang has paid closer attention to the development of AI than most people in politics and he remains a respected commentator on the issue.On “AI Safety”
“A pressing need — right now there are about 100 times more regulations on opening a hot dog stand than there are in launching a new model that might impact millions of people. It’s the Wild West and that’s not a compliment. I’ve heard some leaders call AI ‘the new electricity’ and the power grid is regulated.”
Liberal Believers
Those broadly on the political left or center-left who, unlike the AI skeptics in their political movement, are convinced that AI will be a transformative technology — and that the government should be doing more to think about it.
Alondra Nelson
Professor, Institute for Advanced Study
Place in the cosmos
Nelson served as acting director of the White House Office of Science and Technology Policy under the Biden administration, where she developed the “Blueprint for an AI Bill of Rights” and won the enmity of many AI accelerationists.On “AI Safety”
“AI safety means systems which serve, not harm, real people — the woman denied a mortgage by an inscrutable algorithm, the teenager whose mental health deteriorates from curated feeds, the company whose critical infrastructure is lost to an AI agent’s irreversible error. Fundamentally, it’s about whether AI systems are safe from bias in consequential decisions, safe from manipulation that undermines democratic discourse, safe for young people’s development, safe for workers’ livelihoods. And do they pose risks that could erode the institutions and social fabric holding societies together? These aren’t peripheral concerns — they’re what safety has always meant in our democratic and consumer protection frameworks, demanding that powerful technologies serve people rather than harm them.”
Nat Purser
Director of U.S. policy, AI Verification and Evaluation Research Institute
Place in the cosmos
Purser is a progressive who believes in the power of AI, believes it should be regulated and believes that the Democratic Party should seriously grapple with AI governance.On “AI Safety”
“I like that it implicitly rejects technological determinism — it assumes we can shape how AI develops or gets used. AI will be transformative, but whether that word has a positive or negative valence depends on what we do today.”
Mike Sexton
Senior policy advisor for AI, Third Way
Place in the cosmos
At the center-left group Third Way, Sexton and his organization have, fittingly, often taken something of a middle position in the debate: AI development but with real guardrails.On “AI Safety”
“AI safety is about preventing real harm. That means building a frontier-model vetting system capable of identifying models too risky for public release, and establishing a federal infrastructure with the resources, expertise and agility to monitor emerging threats and respond before they escalate.”
Let It Rip Crowd
Those who believe too much regulation, or the wrong kinds of regulation, will lead to a slowdown in innovation that would set the U.S. back and empower geopolitical rivals, in particular China.
Neil Chilson
Head of AI policy, Abundance Institute
Place in the cosmos
Just as Elmore has a pause sign in her X bio and Soares has a stop sign, Chilson has a range of emojis — a rocket ship, a graph pointing up and various arrows pointing up — making clear that he’s an AI accelerationist.On “AI Safety”
When I hear the phrase ‘AI safety’ my first reaction is ‘safe from what?’ Yes, we want the most advanced computational systems ever developed to be safe and be used safely. But we also want to reduce staggering rates of medical misdiagnosis, decrease traffic fatalities and discover new drugs that support healthier, happier lives. The status quo is quite unsafe. More safety requires progress, not stasis. By democratizing access to expertise, AI tools will help people solve their daily problems and live more prosperous, healthier, and yes, safer lives. AI can also help us tackle ambitious challenges: disease, poverty, war, accidents and the deep mysteries of science, life, intelligence and the cosmos.”
Nathan Leamer
CEO, Fixed Gear Strategies
Place in the cosmos
In addition to his role as a political consultant, Leamer is known as the executive director of Build American AI, the advocacy arm of the super PAC network Leading the Future, which is funded by executives at OpenAI and a16z largely to advocate for candidates and legislation they consider to be pro-innovation.On “AI Safety”
“When I think of the phrase ‘AI safety,’ I think of a fixed mindset. One that obsesses with potential harms and thus hinders their ability to articulate policies that enable growth and opportunity. Safeguards and balance are important, and I see pro-AI voices articulating that in policy discussions. I see a growth mindset manifested in the White House’s AI framework, which is written to enable innovation while mitigating specific harms.”
Paul Lekas
Head of global public policy and government affairs, Software & Information Industry Association
Place in the cosmos
A tech industry lobbyist, Lekas has advocated for soft-touch regulations on AI.On “AI Safety”
“I think there has been some concern that the so-called Doomer community was prioritizing, to a great degree, the existential risks associated with AI without actually looking at the benefits and even the ways that that AI can mitigate some of those risks that were put out there. At the same time, there’s always been talk about the potential of AI on the one side and the risks or the challenges on the other. And so I think that safety is a term that is hard to read outside of that context. I like to talk a lot about reliability. I don’t want to say safety equals reliability, but I think this is what we are ultimately getting at in these different types of efforts. I think at the heart of all this is reliability with respect to model behavior. And all of that ultimately goes to the trust that we have in these systems, because we need to have a degree of trust in them in order to be able to rely on them to do tasks small and large.”
Dean Ball
Head of strategic futures, OpenAI
Place in the cosmos
Ball was the lead author of the Trump administration’s 2025 AI Action Plan and can broadly be described as an accelerationist, even as he has sometimes argued for certain forms of regulation.On “AI Safety”
“It’s actually a term that I really hate, to be honest with you. And the fundamental reason I hate it is because of a great essay that Sayash Kapoor and Arvind Narayanan wrote several years ago now called ‘AI safety is not a model property.’ Many people have attempted to create safety from the top down in AI. In 2022 and 2023, when ChatGPT first came on the screen, oftentimes safety meant ‘how much can we ensure that the model doesn’t say stuff we don’t want it to say.’ We want AI systems that, within reasonable limits, don’t actively contribute to creating egregious new forms of harm in society. That’s basically it. It’s not really about control as it is a modicum of order. And I think safety is a much broader cultural and commercial phenomenon. It’s going to depend immensely on how AI is used, what the laws are surrounding that, what the guardrails are. It’s this broad, huge institutional complex that will actually create safety. And it’s not just some specific property of a model. It’s part of why I’ve always preferred the term governance to safety. Some people would try to rope in ‘AI safety’ with ‘misinformation’ and ‘bias,’ and I think that is actively Orwellian. It can go in all kinds of crazy directions, and you have to reject that entirely. It’s not to say that AI safety isn’t important. It’s to say that it is almost too important to trivialize with making some little ‘trust and safety’ style team. It’s too important to try to hammer down from the top.”