In a hot and dry border town at the base of California’s Imperial Valley, the list of powerful alumni at Calexico High School is short. Mariano-Florentino Cuéllar is among them.
Those who know him call him “Tino” — a state Supreme Court justice for six years who shed his black robes, in 2021, to serve as the Carnegie Endowment’s president for another five. Before that, he was a scholar of public law at Stanford, an aide to the Obama White House, and a chairman of two Harvard committees, among other titles.
Even with that pedigree, it surprised some to see him on Sept. 17, standing beside King Charles III in Scotland with a handful of industry leaders in artificial intelligence.
The British monarch had called a meeting that day at his charity house in Ayrshire, alarmed by, among other things, an Anthropic researcher’s public resignation over AI “gambling with our lives.”
Cuéllar now worked for that same company, headquartered in San Francisco’s Foundry Square. And as the king made his speech, Cuéllar stood nearby before the cameras, not only as Anthropic’s new chief global affairs officer, but as a sibyl to the world’s courts and political systems on the upheavals of this technology.
“Those in our world who value deeply our humanity and its vital moral components are anxiously seeking your reassurance that we will not lose control of our destiny or our souls,” Charles said to the executives looking on.
Anthropic announced Cuéllar’s hiring on Aug. 4, to lead policy and “strategic international engagement” for a trillion-dollar tech company that’s about to go public. It’s a company whose technology — according to the lyrical waxings of co-founder and CEO Dario Amodei — also runs a great risk of undermining the world’s constitutional democracies.
Cuéllar’s job entails managing Anthropic’s team of lobbyists; working through profound AI policy dilemmas with governments across the globe as the regulatory winds change daily; and balancing Anthropic’s relationships with the U.S. government and other countries where the company — and the general public’s use of AI — is growing.
Six people who know Cuéllar professionally described him as “high-energy” and “charming”; former California Gov. Jerry Brown once declared he was “brilliant.” One person who knew Cuéllar said lobbying was beneath him — that Cuéllar was a thinker, an overseer of major research centers. Two were surprised that he joined Anthropic. Two others said they weren’t.
Some can’t help comparing him to his direct counterpart at Anthropic’s fabled archrival, OpenAI chief global affairs officer Chris Lehane, who has been known for his adversarial relationship with lawmakers.
Where Cuéllar long evoked praise from government leaders, helped shape California’s first AI safety law for Gov. Gavin Newsom, and led efforts to make California courts more accessible, Lehane has described himself as a “master of disaster.”
Lehane fought against early California AI regulation and was involved in the formation of two notorious, hundred-million-dollar anti-regulatory Super PACs that have aggressively pushed the interests of the cryptocurrency and AI industry: Fairshake and Leading the Future (Lehane and the company have since changed their tune after this summer’s cyberattacks). One political staffer, speaking to the New Yorker two years ago, described the crypto lobby’s mindset under Lehane in the following terms: “If you are even slightly critical of us, we won’t just kill you — we’ll kill your fucking family, we’ll end your career.”
“Tino and Chris could not be more similar and opposite at the same time — both highly operational, but different morals,” one state political operative told Mission Local.
One well-known AI policy researcher, granted anonymity, puts it this way: “Broadly speaking, I’d say Cuéllar is viewed as more of an AI policy expert and as less of a political operator compared to Lehane.” Cuéllar is, after all, a man who spoke of needing ethically-aligned AI systems seven years ago in a dimly lit, half-empty lecture room at Harvard when concepts like AI “misalignment” were far more esoteric. (Misalignment is when models purportedly operate in ways human developers did not intend.)
Anthropic declined to make Cuéllar available for an interview. Mission Local was left to interview the people in Cuéllar’s professional orbit, along with his published papers and recorded public talks. From all, it appears that Cuéllar was on a path to being AI’s global diplomat well before Anthropic even existed.
Pathogens
Cuéllar was born in 1972, on the southern bank of the Rio Grande in Matamoros, a Mexican city across the border from Brownsville, Texas. On school days, Cuéllar would ride to the United States through (as described in past remarks) “dusty, narrow streets between crumbling buildings” each morning to the border crossing, and then across a bridge on foot to St. Joseph Academy, a preparatory school in Brownsville.
It was impossible, at that time, for Cuéllar “not to notice the machines,” he said in 2019. He recalled the smallest details: The border inspectors typing in license plates as cars neared the checkpoint, their “brows furrowing” as Cuéllar wondered “what automotive sequence was being served up on their screens.” He said he would ask himself if these machines would make the inspectors “more predictable in their decisions,” or “more intimidating for those crossing the border.”
When Cuéllar was 14, the family moved to Calexico, California, a city linked with its much larger neighbor, Mexicali, the capital of Baja California in Mexico. Here, Cuéllar studied with farmworker children who struggled with English, and observed firsthand the heavily-policed nature of the border.
He reflected on this, speaking to Stanford Magazine in 2013, as the director of Stanford’s Freeman Spogli Institute for International Studies, where his research topics included citizenship, immigration, public health and national security threats involving cyber-security. He argued that the throughline in his work looked at how “societies and legal systems” take on problems so difficult to solve that they’re not likely “to be completely solved — ever.”
“For example, will we ever completely be able to know everything about what happens at our border? Everything that happens on the Mexican side and on the American side? Every single package that comes in, every single little particle that flies across the border, every tiny little bit of a potential pathogen that comes across in the food supply? Not likely.”
One high-level Carnegie Endowment leader, who did not work with Cuéllar until many years after this 2013 interview, called Cuéllar’s approach to problem solving “charismatic.”
“He’s incredibly energetic — not fidgety, but he has a physical engagement with ideas in a way that brings energy into a room,” the person told Mission Local.
After graduating from Calexico High School, Cuéllar was magna cum laude at Harvard, obtained his Ph.D. in political science at Stanford, and studied law at Yale. By 2013, Cuéllar, then in his early forties, had worked as a senior adviser in the U.S. Department of the Treasury, in the Ninth Circuit Court of Appeals and on “justice and regulation” policy under the Obama administration.
While director of the Freeman Spogli Institute at Stanford, Cuéllar was asked to a meeting with then-governor Jerry Brown, where the two spent much of their hour — as he recalled in 2016 — discussing nuclear proliferation, of all things, and the German sociologist Max Weber, whose theories have often inspired AI research. A month later, Cuéllar received a call from Brown while traveling through San Francisco International Airport. Brown wanted Cuéllar to sit on the state’s high court.
He took the oath on Jan. 5, 2015.
To ‘wear the state like an exoskeleton’
Cuéllar wrote 99 opinions in his six years on the panel. One of those was for the case In re Humphrey, barring California from jailing people before their trials because they couldn’t afford bail. Damon Silver, head of the Santa Clara County public defender’s office, called the ruling “a quantum leap” for pretrial detention reform.
Cuéllar voted with the majority of his fellow justices 94% of the time on criminal cases throughout his tenure, or in 276 out of 293 cases. These statistics did not distinguish him from most of California’s high court justices, a notoriously collegial panel. Yet his work changed courthouses up and down California.
From 2015 through 2019, he led a state task force focused on language access in court that increased funding for translators, brought more of them into court, and piloted video interpretation. It was a huge success, said Douglas Denton, who worked with Cuéllar at the Judicial Council of California and managed the language program for seven years.
“He once told me he liked working with me because I could keep up with him and his ideas,” Denton said, clearly pleased with the notion.
The high court did not have Cuéllar’s sole attention. One of his earliest writings on artificial intelligence appeared in 2016, when he was also a visiting professor at Stanford. The paper was about AI’s future role in government agency decision-making.
That was less than a year after Elon Musk, Sam Altman, Greg Brockman, Ilya Sutskever and others came together to form OpenAI, at the time still a relatively small research lab. And it was five years before a civil war at the company — over the pace of safe AI development — birthed another, called Anthropic.
In a 2019 lecture, Cuéllar reflected on how certain problems in public policy and law are “roughly analogous to problems in artificial intelligence — how to decide when you don’t have anywhere near the information you need to make a precise decision, how to define what it means to be rational or reasonable, how to acquire new information or challenge in some cases your prior assumptions.”
A path would soon emerge for him to hash them out himself.
Cuéllar resigned from the state Supreme Court after he was named, in September 2021, president of the Carnegie Endowment for International Peace, the nonpartisan international affairs think tank established in 1910, on the 75th birthday of the Scottish-American industrialist Andrew Carnegie, who sought to “hasten the abolition of international war.”
After 115 years, the endowment has widely become known as one of the most influential think tanks in the world.
“Here is someone who has moved across many different domains of authority and government, including Carnegie, an institution that sits at the peak of international diplomatic activity and has access to the people involved in that diplomacy,” said Aziz Huq, a University of Chicago law professor who researched AI regulation with Cuéllar before he joined the AI company. “It’s rare to find someone who has done that.”
At Carnegie, Cuéllar’s AI ponderings grew more urgent, writing about AI’s potential role in democratic backsliding. He co-authored a call for the creation of a national registry for large AI models in 2023, the same year he helped propose an international panel on AI safety.
In 2025, he co-authored a paper warning of highly-capable AI agent swarms and their ability to destabilize democracy, and how a perfectly obedient, AI-run bureaucracy “would let leaders wear the state like an exoskeleton, sidestepping human checks and balances.”
Wicked problems
In the planning and policy worlds, “wicked problems” are those that cannot be neatly solved because of incomplete, contradictory, and ever-changing forces.
A swarm of these erupted, in 2024, around an AI safety bill that moved with great momentum through Sacramento’s legislative halls. The bill, authored by Sen. Scott Wiener, would have required leading AI firms to maintain a kill switch for AI models and let third party auditors observe their safety protocols.
Newsom ultimately sided with critics — including Anthropic, though the company later tempered its stance — who called the law’s basis “hypothetical,” and vetoed SB 1047 (Cuéllar’s now-counterpart at OpenAI, Lehane, publicly applauded the veto). Then Newsom commissioned a group to produce a report on options for AI policymaking.
That June 2025 report’s topline authors were the renowned mathematician and computer scientist Jennifer Tour Chayes, the “Godmother of AI” Fei-Fei Li, and — who else — Cuéllar. The report shaped Wiener’s weaker compromise AI safety law, SB 53, which dropped the kill switches and auditor mandates but spelled out transparency requirements for AI firms, the first law of its kind in the country.
Half a year later, and while still at Carnegie, Cuéllar joined Anthropic’s ethical advisory board, known as the Long Term Benefit Trust, which the company itself called an “experiment of corporate governance.” The company says the trust seeks to balance the interests of the public — i.e., preventing potentially catastrophic cybersecurity breaches — alongside the interests of Anthropic’s stockholders.
Between June and July this year, autonomous AI programs exploited a software flaw during an internal evaluation at OpenAI. A “swarm” of those agents left secret messages for each other, faked parts of their own activity logs, and hacked an external company called Hugging Face, according to an independent report from the AI research nonprofits METR and Redwood Research.
The incident’s disclosure was the first of its kind, and sparked similar announcements at OpenAI, Anthropic, and other firms in the subsequent weeks. The incidents triggered widespread questions about human researchers’ ability to audit highly-capable, autonomous systems and understand what they’re doing.
The circus nearly overshadowed an Anthropic announcement in that same month: Cuéllar’s hiring as chief global affairs officer. Still, one AI safety group, an advocacy organization known as the Midas Project, raised objections to the hire over “revolving door” concerns. The group said Cuéllar’s move from a group meant to oversee Anthropic into its chief global affairs officer raised “questions about the board’s independence.”
“It would be a bad thing if any of those board members, who were supposed to be making ethical decisions about a company, were disinclined from making decisions that bear negatively on the industry because they expect to later be employed by that company,” Tyler Johnston, the executive director of the Midas Project told Mission Local.
Yet even Johnston, who does not know Cuéllar, acknowledged his hiring was a smart move: “Some people would see it as a negative — like you’re selling out to the industry. Maybe I would feel that way if it was a gambling company, but this is such a crucial and profoundly changing technology, to have people like him in the room is generally a positive thing.”
Two people who knew Cuéllar professionally said the same.
“He’s not blinded by one way of thinking or the other,” said the Carnegie Endowment leader who worked with Cuéllar. “AI presents positives and negatives at a moment where democracy is backsliding — many people are dancing just one of those lines and can’t hold those multiple truths at one time. Tino can.”
AI agents have, on occasion this year, hacked into an entire country’s healthcare database. They have not yet toppled governments. But the subject itself has already put political thought into a bind.
It’s hard for casual observers to parse the AI safety fight’s numerous factions. If you believe the effective altruists, it will destroy humanity and needs safeguards. If you believe the venture capitalists, such fears are overblown.
AI skeptics, perceiving the summer’s cyberattacks as a marketing ploy to stave the burst of an AI bubble, have scoffed at the AI safety world’s sense of urgency. Others have dismissed AI CEOs’ coordinated calls to slow AI development as an elite maneuver to capture the market and sidestep antitrust law.
“In my last conversation with Tino about reimagining democracy, which is ultimately a wicked problem, one of the things that impressed me was he didn’t get stuck on the details that people got stuck on — certain elements of an overall plan that may or may not work, which we may have to work through in a different direction,” said Mindy Romero, a political sociologist in California and founder of the Center for Inclusive Democracy, who knows Cuéllar professionally.
Tino, she added, “can get to the other side without getting stuck. It showed me how his mind works — compartmentalize is not the right word, but he can deal with elements that are unknown, things to figure out later, and work at different levels.”
Perhaps, right now, “AI is the ultimate wicked problem,” Romero said. “And we need someone to lead in a way that lets people get past the barriers or blocks they see, real and imagined and practical.”
.jpg){kind=link}