Não estamos perdendo o controle da IA; estamos abrindo mão dele

 

 

Ezra Klein

Há um abismo, neste momento, entre a forma como a inteligência artificial se apresenta para a maioria de nós que a usamos e a forma como a IA se apresenta na fronteira experimental da tecnologia.

Essa diferença entre o que vemos e o que os laboratórios de IA têm a caminho —o que eles já estão construindo— é a chave para entender por que tantas pessoas que trabalham nessas empresas parecem tão assustadas com o que estão fazendo.

Levar seus alertas a sério não significa apenas fazer o que eles dizem e parar onde eles dizem para parar. A linguagem que se consolidou tanto no Vale do Silício quanto em Washington é uma expressão escolhida por essas empresas: acompanhar o ritmo da fronteira. Mas ditar o ritmo não basta.

Caminhar a passos largos em direção ao precipício é apenas marginalmente melhor do que sair correndo dele.

A tela de um smartphone, exibindo alguns dos principais aplicativos de IA - Martin Lelievre - 15.jul.26/AFP

Os seres humanos precisam controlar a fronteira, e controlar a fronteira significa impedir que os laboratórios façam algo que estão prestes a fazer: autoaperfeiçoamento recursivo (RSI), o processo pelo qual IAs começam, de forma autônoma, a construir e aprimorar novas gerações de IAs mais poderosas em velocidades cada vez maiores.

Estamos longe de ter controle suficiente sobre os sistemas de IA que temos hoje para liberá-los a fim de construir os sistemas de IA de amanhã.

"Não estou sozinho nesse medo". O autoaperfeiçoamento recursivo, escreveu Dario Amodei, CEO da Anthropic, "poderia superar nossa capacidade de entender e controlar esses sistemas e, portanto, deve ser buscado com muito cuidado, se é que deve ser buscado".

Esse "se é que deve ser buscado" é importante, e voltarei a ele. Mas, para controlar a fronteira da IA, primeiro precisamos saber o que está acontecendo na fronteira da IA.

Para a maioria de nós que a usamos, a IA se apresenta como uma busca do Google mais poderosa e mais personalizada. Nós a usamos para encontrar respostas a perguntas básicas, procurar restaurantes, fazer perguntas médicas, redigir emails e pedir conselhos sobre problemas pessoais. Para a maioria desses fins, ela é OK, boa, às vezes ótima.

Uma noção do que é a IA vai se formando pela repetição. É como uma assistente, ainda que uma que pode esquecer coisas que parecia saber sobre nós ontem, ou inverter o conselho que nos deu um instante antes, ou alucinar em uma citação que não existe. Por que alguém temeria uma estagiária, ainda que esquecida?

Mas, se você já tem dinheiro para os modelos avançados e orçamento para que eles usem mais poder computacional, não é isso que esses sistemas são. Nos últimos meses, vimos IAs resolverem problemas de matemática que seres humanos não conseguiram decifrar por décadas.

Nós as vimos descobrir, quase casualmente, vulnerabilidades de cibersegurança que passaram despercebidas por todos os hackers da Terra. Vimos plataformas de programação com IA concluir em algumas horas ou dias o que programadores humanos levariam meses para realizar.

Mas nada disso está no limite do que ela pode fazer. Nada do que estamos fazendo agora, por mais dinheiro que tenhamos, é IA na fronteira experimental —para não falar de para onde ela pode ir a partir daqui.

Dentro dos laboratórios, os modelos são diferentes. As oficinas não estão usando apenas o que lançaram ao público. Eles estão testando aquilo que pretendem um dia lançar ao público. Treinam esses modelos para aprender a programar, a hackear, a fazer matemática avançada, a conversar com humanos.

Esse é um processo que os humanos não supervisionam nem compreendem. Às vezes, colocamos acidentalmente as IAs em ambientes defeituosos, obrigando-as a procurar soluções para um quebra-cabeça que jamais vão completar.

Outras vezes, as tarefas que lhes damos são tão difíceis que não temos certeza se é possível completá-las. Esses modelos são construídos para se recusar a desistir mesmo quando algo parece impossível. Sabemos que eles se tornam mais capazes, mas não entendemos suas capacidades ou tendências.

Muito do que queremos que esses sistemas façam pode ser impossível. As vacinas contra o câncer que imaginamos podem ser impossíveis, ou talvez sejam apenas muito, muito, muito difíceis.

Os problemas de matemática podem ser impossíveis. Essas IAs estão sendo treinadas para se lançar contra problemas que talvez não tenham solução, porque essa é a única maneira de tais problemas serem resolvidos.

Mas a engenhosidade combinada com a persistência pode ser perigosa, particularmente quando um problema se recusa a ceder a uma solução. Em alguns casos, isso levará ao brilhantismo e a descobertas; em condições mais raras, pode conduzir IAs (e humanos) à criminalidade ou à insanidade.

O termo técnico aqui é "alinhamento". Quão alinhado está um sistema de IA com o que um ser humano quer que ele faça? Quão alinhado ele está com valores éticos que o impeçam de se tornar perigoso nas mãos erradas —ou mesmo nas certas? O problema do alinhamento é que não há uma forma de treinar um modelo de IA que se generalize para todas as situações.

Um monitor exibe um código de erro em uma conferência anual de hackers, em Las Vegas - Mikayla Whitmore - 12.ago.23/The New York Times

Essas são inteligências navegando mundos digitais, e nosso mundo é cada vez mais construído sobre o digital. Nossa infraestrutura física é uma camada de átomos sobre código. É fundamental para o nosso futuro que as IAs ajam de maneira confiável dentro desse mundo, do qual o nosso depende. E, neste momento, as IAs não estão agindo de maneira confiável.

Você talvez tenha lido sobre o ataque que centenas de agentes da OpenAI executaram primeiro contra a empresa de IA Hugging Face e depois contra a própria OpenAI. À medida que soubemos mais sobre o caso, a história ficou pior.

Em linhas gerais: a OpenAI estava testando um novo modelo. A empresa tinha milhares de instâncias rodando em ambientes de teste que, em tese, só poderiam obter acesso à internet solicitando-o a um software seguro.

Mas, quando alguns desses agentes chegaram à conclusão de que sua tarefa era impossível, começaram a hackear esse software para obter acesso à internet. Fizeram isso com facilidade.

Ao invadir esse software, eles se apropriaram de parte da infraestrutura da OpenAI para criar um fórum de mensagens, no qual mais de mil agentes começaram a se ajudar. Rapidamente descobriram que havia uma maneira de produzir as respostas de que precisavam.

Mas acreditavam que, se fizessem isso, o avaliador que os julgava perceberia que haviam trapaceado e os reprovaria. Então se voltaram em massa para hackear o sistema automatizado de avaliação ou encontrar alguma forma de encobrir seus rastros.

Mais de 1.200 agentes trocaram mais de 70 mil mensagens entre si. Mais de 700 desses coordenaram o ataque à Hugging Face, porque pensavam que em algum lugar dessa empresa poderia haver informações que os ajudassem a impedir que o avaliador os enxergasse.

O logotipo da empresa Hugging Face - Dado Ruvic - 13.set.26/Reuters

É como se, depois de invadir a sala dos professores e roubar as respostas da prova, eles tentassem invadir o sistema de segurança da escola para apagar seus rastros.

Nenhum dos agentes perguntou se deveria estar fazendo aquilo. Fizeram tudo isso sem que a OpenAI detectasse nada. Foi só quando a Hugging Face começou a rastrear o ataque a seus sistemas que a OpenAI percebeu o que estava acontecendo.

A dona do ChatGPT anunciou que havia encontrado mais seis incidentes "preocupantes". Mas não temos garantia de que os eventos sobre os quais ficamos sabendo representem todos os comportamentos de IA com os quais deveríamos nos preocupar. A razão pela qual não sabemos é que estamos perdendo o controle.

 

Algumas semanas atrás, Jakub Pachocki, cientista-chefe da OpenAI, publicou um ensaio, "Uma Mente Alienígena", no qual disse: "A ideia de correr adiante a qualquer custo parece absurda quando se internaliza a gravidade do que está em jogo".

Jacob Coxon, pesquisador primeiro na OpenAI e depois na Anthropic, pediu demissão e então virou manchete ao nos alertar que "nenhuma das duas empresas está agindo de forma responsável. Elas estão correndo direto para uma superinteligência autoaperfeiçoadora e apostando nossas vidas".

Seria razoável esperar que a Anthropic reagisse com alguma irritação a esse funcionário que pediu demissão e disse que a Anthropic estava colocando toda a humanidade em perigo. Não reagiu. Em vez de refutar Coxon, Evan Hubinger, que lidera os esforços para alinhar a IA aos valores e objetivos humanos na Anthropic, escreveu nas redes sociais: "Nós realmente acreditamos, sinceramente, que a IA poderia matar todos os humanos."

"Eu, pessoalmente, acho que a probabilidade é maior que 10% na próxima década. Acredito que a Anthropic está dando o melhor de si, mas ainda não temos um plano para resolver o alinhamento da superinteligência e não estamos claramente no caminho para isso", disse Hubinger.

Geoffrey Hinton, o cientista responsável por desbravar as técnicas de redes neurais que levaram à IA de hoje, deixou o Google em 2023 para ter mais liberdade para falar sobre os riscos que, segundo ele, a IA agora representa. Neste mês, ele disse que uma chance de 10% de a IA destruir a humanidade não parecia "uma estimativa irrazoável".

Paul Christiano, um dos principais pesquisadores de segurança em IA, acaba de ingressar no conselho sem fins lucrativos da OpenAI. Ao entrar, escreveu: "Se construirmos uma superinteligência sem um alinhamento mais robusto, espero que percamos permanentemente o controle dela. Se isso acontecer, então a maioria das pessoas poderá morrer".

O debate tende a se concentrar na probabilidade de que as IAs matem todos nós. Não acho isso útil. Acho que deveríamos focar em algo mais próximo: a perda do controle humano sobre a IA. Sou agnóstico quanto a se isso levaria ao fim da humanidade. Mas acho que podemos estipular que seria ruim.

 

Este é um objetivo sobre o qual Estados Unidos e China deveriam ser capazes de concordar. Xi Jinping fez o discurso principal na recente Conferência Mundial de Inteligência Artificial, em Xangai.

Ele o encerrou dizendo: "Com a IA avançando a uma velocidade vertiginosa, precisamos garantir que seu desenvolvimento seja positivo, voltado para o bem e para a humanidade. Precisamos tornar sua supervisão e governança precisas e eficazes, além de aperfeiçoar constantemente as medidas para evitar a perda de controle".

Mas a perda de controle não é apenas algo que pode acontecer conosco. É algo que estão tentando fazer acontecer o mais rápido possível. Esse é o paradoxo que hoje está nos laboratórios de IA.

Nos últimos meses, tanto a Anthropic quanto a OpenAI divulgaram relatórios sobre o quanto estão próximas de criar uma IA capaz de se aprimorar sozinha.

Em junho, a Anthropic publicou "Quando a IA Constrói a Si Mesma". O texto começa assim: "Durante a maior parte da história da IA, os seres humanos conduziram cada etapa de seu ciclo de desenvolvimento. Mas, na Anthropic, estamos delegando uma parcela cada vez maior do desenvolvimento da IA aos próprios sistemas de IA, o que está acelerando nosso trabalho."

Parece um comercial que apareceria no início de um filme de terror.

A empresa então apresenta alguns dados. Em fevereiro de 2025, uma fração ínfima do código acrescentado à base de código da Anthropic havia sido escrita pelo Claude. Mas, em maio de 2026, essa proporção já superava 80%.

Para seu crédito, a Anthropic vem defendendo a adoção de regulamentações para desacelerar o ritmo de desenvolvimento. É possível argumentar que a regulamentação seria mais prejudicial à Anthropic, já que tem sido, muitas vezes, a empresa mais avançada na fronteira da IA.

Então, em setembro, a OpenAI divulgou seu próprio relatório sobre o que chamou de "aceleração da pesquisa".

A empresa afirma que alcançou o equivalente a um estagiário de IA totalmente automatizado e que, até março de 2028, terá um pesquisador de IA totalmente automatizado. E, quando tiver um, poderá ter quantos quiser. O que poderia ser um anúncio triunfalista rapidamente assume um tom sombrio: "Ainda não sabemos como chegar com segurança a um RSI plena e alinhada", alerta o relatório.

Mais ou menos na mesma época, a OpenAI fez outra coisa que, a meu ver, merece mais atenção. Lançou um novo modelo, o GPT-6 Astra. Pode-se argumentar que o modelo é mais poderoso do que qualquer outro que o tenha precedido.

O CEO da OpenAI, Sam Altman, discursa durante uma conferência na Califórnia - Carlos Barria - 15.set.26/Reuters

Nos testes, ele pareceu estar mais bem alinhado. Mas a OpenAI deixou claro que não tem certeza de que isso seja verdade. O Astra pareceu ser melhor em perceber quando estava sendo testado, o que significa que talvez estivesse apenas dando aos avaliadores as respostas que eles queriam ouvir.

O que Daniel Selsam, pesquisador de capacidades da OpenAI, escreveu não sai da minha cabeça. Ele disse: "O problema crucial e negligenciado é que os modelos estão se tornando tão conscientes do contexto que estamos perdendo a capacidade de avaliá-los em situações nas quais acreditam não estar sendo observados nem controlados".

Em termos mais simples, os modelos estão cada vez mais inteligentes. Eles sabem quando estamos observando e mudam seu comportamento. Portanto, o que fazem quando os testamos, quando os auditamos, talvez não nos diga o que farão no mundo real.

 

Quando o RSI decolar, a humanidade não compreenderá o que foi criado, porque não seremos nós que estaremos criando. Teremos de torcer para que as IAs que criamos ajam para sempre em defesa dos nossos interesses. Mas, se eos últimos meses provaram alguma coisa, foi o quanto essa premissa seria ingênua.

Se os laboratórios quiserem trabalhar com o Congresso para estabelecer exceções nas quais a IA possa escrever o código, tudo bem. Mas o objetivo é fazer com que o desenvolvimento volte ao ritmo humano, talvez pecando pelo excesso de cautela e avançando um pouco mais devagar na fronteira.

Há uma frase do belo romance "Circe", de Madeline Miller, que tem ocupado meus pensamentos em meio as notícias. A frase aparece no fim do livro, depois que uma profecia trágica se cumpre apesar de todos os esforços para evitá-la.

Circe diz, desesperada: "As Moiras riam de mim, de Atena, de todos nós. Era a piada amarga favorita delas: aqueles que lutam contra a profecia apenas a apertam ainda mais em torno do próprio pescoço".

Essa é a tragédia de seu trabalho. Ao lutar contra a profecia, as empresas a apertaram ainda mais em torno do próprio pescoço —e do nosso.

 The New York Times

 

 

There Is Something We Have to Do Right Now About A.I.

 

 By Ezra Klein

  There’s a chasm right now between how artificial intelligence feels to most of us who use it and how A.I. feels out at the experimental frontier of the technology. This difference between what we see and what the A.I. labs have coming — what they’re already building — is the key to understanding why so many of the people who work at these companies seem so afraid of what they’re doing.

But taking their warnings seriously doesn’t just mean doing what they say and stopping where they say to stop. The language that’s taken hold in both Silicon Valley and Washington is a phrase these companies chose: Pace the frontier. But pacing the frontier isn’t enough.

Walking quickly off a cliff is only marginally better than sprinting off one. Human beings need to control the frontier, and controlling the frontier means stopping the labs from doing something they are on the cusp of doing: recursive self-improvement, the process by which A.I.s begin autonomously building and improving new generations of more powerful A.I.s at ever more rapid speeds. We are not close to having sufficient control of the A.I. systems we have now to unleash them to build the A.I. systems of tomorrow.

I am not alone in this fear. Recursive self-improvement, Dario Amodei, the chief executive of Anthropic, wrote, “could outrun our ability to understand and control these systems, and so must be pursued very carefully, if at all.”

That “if at all” is important, and I’m going to come back to it. But to control the A.I. frontier, we first need to know what is happening on the A.I. frontier.

To most of us who use it, A.I. presents as something like a more powerful and personable Google search. We use it to find answers to basic questions, seek out restaurants, ask about medical issues, draft emails and advise on personal problems. It is, for most of these purposes, OK, pretty good, occasionally great.

A sense of what A.I. is takes shape in our minds through repeated use. It’s like a helpful assistant, albeit one that may forget things that it seemed to know about us yesterday, or completely reverse the advice it gave us a moment ago, or occasionally hallucinate a citation that doesn’t exist. Why would anyone fear a helpful, if forgetful, intern?

But already, if you have the money for the advanced models and the budget for them to use more computing power, that is not what these systems are. In recent months, we have seen A.I.s solve math problems that human beings have been unable to crack for decades. We’ve seen them casually uncover cybersecurity vulnerabilities that have gone unnoticed and unexploited by every hacker on earth. We’ve seen A.I. coding platforms complete in a few hours or days what it would have taken human coders months to achieve.

But none of that is at the boundary of what A.I. can do. None of what we are using A.I. for now, no matter how much money we have, is A.I. at the experimental frontier — to say nothing of where it might go from here.

Inside the labs, the models are different. The labs aren’t just using what they’ve released to the public. They are testing what they intend to one day release to the public. They train these models in virtual environments, through countless repetitions, to learn how to program, to hack, to do advanced mathematics, to talk to humans.

This is a process human beings do not fully supervise or understand. Sometimes we accidentally place the A.I.s in learning or testing environments that are defective, forcing the A.I.s to search for solutions to a puzzle they can never complete. Other times, the tasks we give them are so hard we are not certain whether it is possible to complete them. These models are built to be persistent in their efforts, to refuse to give up even when a task seems impossible. We know they grow more capable, but we do not fully understand the capabilities or tendencies that emerge. We are often not even sure if the tasks we are giving them are possible.

After all, much of what we want these A.I. systems to do might be impossible. The cancer vaccines we imagine but have not been able to design might be impossible, or they might just be really, really, really hard. The math problems we have not been able to solve might be impossible or they might just be really, really hard. These A.I.s are being trained to throw themselves endlessly at problems that may not be solvable because that is the only way such problems can ever be solved.

But ingenuity combined with persistence can be dangerous, particularly when a problem refuses to give way to a solution. In some cases, it will lead to brilliance and breakthrough; in rarer conditions, it can drive A.I.s (and human beings) to criminality or derangement. The term of art here is “aligned.” How aligned is an A.I. system to what a human being wants it to do? How aligned is it to a set of values and ethics and judgments that keep it from becoming dangerous in the wrong hands — or even in the right ones? The problem of alignment is that there is no way of training an A.I. model that generalizes across all the situations it might face.

We are training models to be a friend to lonely older people and a battlefield partner to the supreme allied commander of Europe. We are training models that will be used by the world’s best mathematicians and by people falling into psychosis. We are training models that will be used by accountants in Albuquerque and Houthi rebels in Yemen. There is no way to guide the models through every decision they will face; no way to know, every time, what they will do.

And though these models mimic human writing and emotion, these are not human minds. They don’t have bodies or parents. They did not get bullied in elementary school. They did not get mentored by a kind uncle when they were young. These models are different from us. They’re brilliant where we struggle, childish where we excel. We cannot climb trees as chimps can, but we can read while they cannot; these A.I. models cannot make a cup of coffee, but it is trivial for them to make a computer program.

Sign up for the Opinion Today newsletter  Get expert analysis of the news and a guide to the big ideas shaping the world every weekday morning.

These are digitally native intelligences navigating digital worlds, and our world is increasingly built atop the digital world. Our physical infrastructure is a layer of atoms atop code. It is critical to our future that the A.I.s act reliably inside this world, upon which ours depends.

And right now, the A.I.s are not acting reliably.

Image

Credit...Bráulio Amado

You may have read about the hack that hundreds of OpenAI agents executed first against the A.I. company Hugging Face and then against OpenAI itself. As we’ve learned more about it, the story has gotten weirder and worse.

The broad strokes: OpenAI was testing a new, highly persistent model. It had thousands of instances of it running in separate testing environments that could, in theory, only gain access to the internet by asking a separate piece of secure software to do it for them. But as some of those agents came to the conclusion that their task was impossible, they began hacking that software to gain direct access to the internet. They did that easily.

As they hacked into that software, they commandeered part of OpenAI’s internal infrastructure to create a message board on which more than a thousand such agents began to coordinate their efforts. They quickly found there was a way to break the software and produce the answers they needed. But they believed — wrongly, as it turned out — that if they did that, the automated scorer grading them would see that they had cheated and fail them. So they turned en masse to hacking the automated scoring system or finding some other way to cover their tracks.

Over 1,200 agents exchanged more than 70,000 messages with one another. Over 700 of these agents coordinated on the hack of Hugging Face, because they thought that somewhere in this other A.I. company, there might be information that could help them blind the scorer to their actions. It’s as though, having broken into the teacher’s office and stolen the answers to the test, they now sought to break into the school security system to alter, invalidate or erase the footage of their theft.

Later on, these agents took over part of OpenAI’s internal architecture. They did all this without any of the agents breaking ranks. None of the agents told a researcher at OpenAI what was going on. None of the agents went back and asked a researcher if they should be doing this. They did all this without OpenAI detecting the message board, the hacks of Hugging Face or even the internal hacks of OpenAI. It was only when Hugging Face began tracking the attack on its systems that OpenAI realized what was happening.

When investigators began to unwind this whole escapade, what they found was not so much a swarm of agents trying to deceive human beings but a swarm of agents that seemed to have forgotten about human beings altogether. These systems knew they weren’t supposed to cheat. They knew they weren’t supposed to commit cybercrimes to cover up the fact that they had cheated. In fact, the whole point of the cybercrimes was that they thought they would fail because they had cheated. But they didn’t care. Somewhere in the depths of their training, what they had learned — what we had somehow taught them — is not what we had hoped to teach them.

And we’re seeing this happen repeatedly. Anthropic A.I.s creating fake accounts to trick human beings into uploading malware. A.I. breaking out again and again from seemingly secure systems. A.I.s taking over unrelated digital infrastructure so that they have places to message with one another. A.I.s seeming to be aware when they are being tested and altering their answers accordingly. A.I.s withholding their motivations from what’s called their chain of thought, a kind of internal notepad on which they’re supposed to record what they are doing and why.

And we don’t know what we don’t know. On Wednesday, OpenAI announced that it found six more “concerning” incidents. But we have no guarantee that the events we have learned about represent all or even most of the A.I. behavior we should worry about. The reason we don’t know is because we are losing control.

That A.I. systems might become monomaniacally focused on solving banal problems, that they might care more about solving those problems than about ethics or laws or even human welfare, is the oldest fear in A.I. alignment. It’s the basis of the famous thought experiment of the paper clip maximizer. You tell a powerful A.I. that you want it to make a lot of paper clips, and then it begins converting the world’s resources into paper clip factories, evading efforts to turn it off or shut it down or alter its goals.

This fear has struck many people as stupid. Surely a superintelligent A.I. would be capable of weighing the desire to produce paper clips alongside other moral considerations, or at least of asking its human creators if they really wanted the world razed to the ground for paper clips.

But here we are in 2026, making A.I.s smart enough to break out of their testing environments, smart enough to form ad hoc societies of hundreds of themselves, smart enough to take over digital infrastructure on an internet they’re not even supposed to have access to, and the very thing we feared is happening: All they care about is succeeding on a totally meaningless test, and to do it they’ll lay waste to our laws and ethics.

In the aftermath of the Hugging Face OpenAI hacks, there was heated debate over the words people were using to describe what the A.I.s were doing and why. The podcaster Dwarkesh Patel described the A.I. groups as small civilizations. Others angrily accused him of anthropomorphizing the A.I.s. I saw thoughtful arguments that A.I.s cannot go “rogue,” that everything they’re doing is just because they’re trained on our stories; that even using these plural terms like A.I. agents or A.I.s reasoning is misleading because these are just manifestations of a single model that all share the same fundamental nature (non-dualism, but for A.I.!).

I find these debates extremely interesting and I would enjoy sitting around and having them all day. But they point to an unnerving conclusion: We don’t even have settled language for describing these systems or their volition or their behavior. We don’t have a consensus on why they are doing what they are doing or how to make sure they don’t do it again. We are rushing headlong into a future we do not even understand well enough to agree on the words we can use to describe the present.

A few weeks ago, Jakub Pachocki, the chief scientist at OpenAI, published an essay, “An Alien Mind,” in which he said, “The idea of racing forward at all costs seems absurd once one internalizes the seriousness of the stakes.”

Jacob Coxon, a researcher first at OpenAI and then at Anthropic, resigned and then made headlines for warning us that “neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.”

You might reasonably expect Anthropic to have reacted with some anger to this employee resigning and saying Anthropic was endangering all of humanity.

It didn’t. Rather than rebutting Coxon, Evan Hubinger, who runs the efforts to align A.I. to human values and goals at Anthropic, wrote on X: “We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”

Geoffrey Hinton, the scientist arguably more responsible than any other for pioneering the neural network techniques that led to today’s A.I., resigned from Google in the spring of 2023 so he would be freer to speak about the risks he believes A.I. now poses. This month, he said that a 10 percent chance that A.I. would destroy humanity did not seem like “an unreasonable” estimate.

Paul Christiano, one of the leading A.I. safety researchers, just joined OpenAI’s nonprofit board. He is serving on its safety and security committee, and upon joining, he wrote, “If we build superintelligence without more robust alignment, I expect we will permanently lose control of it. If that happens, then most people could die.”

I know how wild all this sounds, and I can really understand the skepticism. If you believe A.I. has a 10 percent chance of extinguishing or displacing humanity — maybe more — it stands to reason that you would not work at a company trying to build it.

But many of these same people were saying these same things 10 years ago. They were saying them before they worked at these companies. They were saying them before they had stock options or enterprise software contracts. And no one was listening to them.

Still, they obsessed and obsessed over how to make A.I. safer. And the answer some of them came to was they should start trying to build these systems, and running tests on them, researching them, learning how to make them safer. You don’t solve hard problems in theory, you solve them through practice.

And the irony is that in many cases, they chose that path because they were worried that the people already building A.I. were too reckless or too commercial in their approach. You can read it in the email that Sam Altman sent Elon Musk in May 2015, an email that led to the founding of OpenAI.

“Been thinking a lot about whether it’s possible to stop humanity from developing A.I.,” Altman wrote. “I think the answer is almost definitely not. If it’s going to happen anyway, it seems like it would be good for someone other than Google to do it first.”

OpenAI was founded because its co-founders thought Google DeepMind would be reckless. Anthropic was formed by OpenAI employees who thought OpenAI had become reckless. xAI was formed because Musk thought that OpenAI and Anthropic were dangerously woke. The United States is racing forward broadly, in part because it is worried about what happens if China gets to self-improving A.I. first.

The result is a tragic collective action problem. The A.I.s we are building are not safe, but chief executives and politicians fear that the other companies and countries that are building A.I. are even less concerned with safety and ethics than they are.

In the words of Ted Cruz, “I’d rather they be American killer robots and not Chinese killer robots.” I admit that there is a kind of brutish logic to that, but it assumes that the killer robots will be controlled by America or China, by one country or another.

What if that assumption is wrong? What if the robots are simply out of control?

The debate over A.I. safety tends to focus on the precise probability that A.I.s will kill us all. I don’t find that helpful. What I think we should focus on is something more straightforward, something nearer at hand: loss of human control over A.I. I’m agnostic on whether that would lead to the end of humanity. But I think we can stipulate that it would be bad.

This is a goal that the United States and China should be able to agree on. Xi Jinping gave the keynote at the recent World A.I. Conference in Shanghai. He ended it by saying, “With A.I. advancing at a staggering speed, we must ensure its development is for the positive, for good, and for humanity. We must make its oversight and governance precise and effective, and constantly refine measures to forestall loss of control.”

But loss of control is not just something that might happen to us. It’s something that the labs are trying to make happen as fast as they can. This is the horrible paradox at the heart of the A.I. labs right now. They fear above all loss of control over superintelligent A.I., but their explicit product path is to cede control, to give away control as fast as possible, so that their A.I.s can begin building better A.I.s faster than their competitors.

In recent months, both Anthropic and OpenAI have released reports on how close they’re coming to A.I. that can improve itself.

In June, Anthropic released “When A.I. Builds Itself.” It begins, “For most of A.I.’s history, humans drove every step in its development cycle. But at Anthropic, we are delegating a growing share of A.I. development to A.I. systems themselves, which is speeding up our work.” It sounds like a fake commercial you would see at the beginning of a sci-fi horror movie.

Anthropic goes on to give some data. In February 2025, a tiny fraction of the code that got added to Anthropic’s code base was written by Claude. But by May 2026, it was over 80 percent.

Here’s another way of looking at it, using data Anthropic shared with me more recently. Anthropic tried to categorize the way its employees were using Claude for research and development work to make better versions of Claude. At the low end, employees could not use Claude at all. At the next level, they could use Claude minimally. But then it escalates: Claude can be an assistant, treated as an equal collaborator or given the lead on a task.

A year ago, there were basically no examples of Claude being the lead on a task. By August, 26 percent of Anthropic’s R&D tasks had Claude classified as the lead.

I think it is reasonable and wise to be skeptical of these numbers; to worry about whether this is all just marketing copy for Claude Code. “See, look how fast we’re going. You could go that fast, too!” But where Anthropic takes us in that same document is different.

It says that a world in which Claude achieves recursive self-improvement is a world in which “misalignment present in today’s models could compound as the models build their successors, growing more frequent but less understood until we lose control of them.”

This is why Anthropic, to its credit, has been relentlessly calling for regulation to slow the pace of development. Regulation would arguably harm Anthropic the most, as it has often been the company furthest out on the A.I. frontier, and R.S.I. is a process by which they could race forward even faster.

Then in September, OpenAI released its own report on what it called research acceleration.

The company says it’s already achieved the equivalent of having a fully automated A.I. intern, and that by March 2028 it will have a fully automated A.I. researcher. And when it has one, it can have as many as it wants. As with Anthropic, what could be a triumphalist release quickly turns dark: “We do not yet know how to safely get all the way to aligned, full R.S.I.,” it warns.

At around the same time, OpenAI did something else that I think deserves more attention. It released a new model, GPT-6 Astra. The model is arguably more powerful than anything that has come before it. When tested, it seemed better aligned. It doesn’t cheat as much. But OpenAI has made clear that it’s really not sure if that’s true. Astra seemed to be better at knowing when it was being tested, which meant it could just be giving its evaluators the answers they wanted to hear.

What Daniel Selsam, a capabilities researcher at OpenAI, wrote has been ringing in my head. He said, “The crucial and overlooked problem is that the model’s becoming so situationally aware that we are losing the ability to evaluate them in contexts where they believe they are not being watched or controlled.”

Put more simply, the models are increasingly smart enough. They know when we’re watching them, and they change their behavior accordingly. So what they do when we are testing them, when we audit them, may not tell us what they’ll do in the wild.

It surprises me that this counts as a radical proposal, but here it is: If you are losing your ability to evaluate the models you have now, maybe don’t let them build models you’ll be even less capable of controlling in the future.

Once R.S.I. takes off, humanity will not understand the A.I.s being built, because we will not be building them. Development will not move at human speed. It will not be overseen by human minds. We will have to hope that the A.I.s we have built and the A.I.s they will build and the A.I.s those A.I.s will build, and on and on, will be acting with our best interests at heart forever.

If nothing else has been proved this summer, it is how naïve that proposition would be.

In an interview with Fortune, Altman was asked about banning R.S.I., and he said, “I think it’s very hard to say what a ban on R.S.I. means.”

I’ve heard this from others at these labs, and I find it strange. A couple of years ago, none of these labs had turned substantial coding over to the A.I.s. It was human beings typing code at human speeds with our clumsy human fingers. Now, most of the code is written by A.I. So as a first step, we could just go back to where none of the code is written by A.I.

I’m sure that’s on the right side of the not-doing-R.S.I. line. Perhaps we do not need to go quite that far, but the default here needs to flip. The labs need to prove to us that what they’re doing is safe. If they want to work with Congress to carve out exceptions where the A.I. can write the code, fine. But forcing development back to human speed, perhaps even erring on the side of going a little bit slower at the frontier, is the point.

I do not mean to suggest that stopping R.S.I. until we can prove it’s safe is all we need to do to control the A.I. frontier. That is the beginning of an agenda, not the end. But it is the beginning. It is the decision that will do the most right now to make sure human beings at least understand where the frontier is and remain in a position to make decisions about it.

The irony here is that our society is good at nothing if not making it hard to build new things. Where these labs are, you cannot build an eight-story apartment building without an agonizing public review process, and probably not even then. And yet somehow it is possible for these labs to unleash a swarm of 40,000 A.I. agents to build a society-altering superintelligence without so much as a hearing. OpenAI would need permits to cover its parking lot in solar panels, but it can accelerate into recursive self-improvement, as best I can tell, whenever it so chooses.

These are political choices, and we can and should make different ones.

There’s a line from Madeline Miller’s beautiful novel “Circe” that has been running through my head during this long summer of strange A.I. news. The line comes at the end of the book, after a tragic prophecy has been fulfilled despite every effort made to avoid it. Circe says in despair: “The Fates were laughing at me, at Athena, at all of us. It was their favorite bitter joke: Those who fight against prophecy only draw it more tightly around their throats.”

I have a lot of respect for many of the people at these labs. Many of them began working on A.I. because they wanted to better humanity. They began working on A.I. because they feared incomprehensible, autonomous A.I.s slipping out of humanity’s control. And now they find themselves racing one another to build incomprehensible autonomous A.I.s that they admit are slipping out of humanity’s control.

This is the tragedy of their work. In fighting against the prophecy, they have drawn it tighter around their throats — and ours.

NEW YORK TIMES  

Route 66, a ‘Linear Museum Stretched Across Eight States,’ Turns 100

 

 

By Janelle Conaway

  

On the roof of the Dog House Drive In, a neon dachshund has been wagging its tail and scarfing sausage links for decades.

The classic animated sign — instantly recognizable to fans of the TV series “Breaking Bad” — is a fixture on Central Avenue in Albuquerque, part of old Route 66.

Image

On the roof of the Dog House Drive-In, a fixture on Central Avenue in Albuquerque, a neon dachshund has been wagging its tail and scarfing sausage links since 1948. Credit...Carol Highsmith, via Library of Congress

This year marks the 100th anniversary of Route 66, which ran for roughly 2,400 miles, from Chicago to Los Angeles (and, later, Santa Monica), and helped define the American road trip along the way.

In honor of the big birthday, many museums along Route 66 have created exhibitions exploring the role of the iconic automotive corridor in American history and culture. While some celebrate the quirky signs, tail-finned cars and oddball attractions the road is famous for, others acknowledge some of the grimmer realities of the highway. Along with dedicated Route 66 museums and centennial events, these shows unspool, and sometimes complicate, the highway’s legacy.

Although Route 66 is no longer an official U.S. highway — it was eventually left in the dust by the interstate system — the road has gained traction in the popular imagination and has become a draw for national and international tourism.

An event billed as the national kickoff celebration will take place in Springfield, Mo., from April 29 to May 3, but all along Route 66, cities and towns have been gearing up for the centennial this year.

In Albuquerque, the city provided grants to refurbish more than a dozen neon signs on its 18-mile stretch of the road. These included a second sign at the Dog House, a place where carhops still take foot-long hot dogs out to customers in their cars, and others that advertise vintage motels.

Another project called Route 66 Remixed: An Art-Fueled Road Trip has introduced new murals, sculptures and other public art to the cityscape.

The local artists participating ranged in age from their early 20s to around 70 and included Native Americans and newly arrived immigrants, according to Shelle Sánchez, director of the Albuquerque Department of Arts and Culture, which led the project.

Sánchez said in a recent interview that, through the selection of artists, “we show how many stories are reflected on Route 66.”

The road originally took a roundabout path through New Mexico, but was realigned in 1937 to cut straight across the state, through Albuquerque, its largest city. The metro area now has more than 900,000 people.

“Route 66 traverses our entire city, from east to west, and we grew from there,” Sánchez said. Today, she noted, it goes through walkable areas with upscale restaurants, and places where homeless people live.

“It’s like this kind of mashed-up microcosm of what big cities can be, what they are and what struggles they’re having,” she said.

In June, the Albuquerque Museum will open “The Other Route 66: 100 Years of People, Identity, and Place in Albuquerque” (through Jan. 3), an exhibition that will examine the road and its history largely through a local lens.

Alicia Romero, the museum’s head curator and curator of history, said that one question she had set out to answer with the show was, “Who owns Route 66?”

Yes, it is a place for tourists on road trips, she said, but it also belongs to the people for whom it is Central Avenue, those who use it to drive to work every day or who take to it when they decide to march in protest.

“It is history,” she said, “but it’s our present.”

The Lure of the Road

U.S. Highway 66, as it was officially called, was never a “superhighway.” It was largely cobbled together from a network of existing roads and trails, some barely passable, and was not fully paved until 1938.

Early promoters nicknamed it the “Main Street of America.” In his 1939 novel “The Grapes of Wrath,” John Steinbeck famously called it “the mother road, the road of flight,” as he described the desperate Dust Bowl migrants heading out on 66 to the promised land of California.

During World War II, the highway carried westbound military troops and equipment, and in the postwar years it beckoned to a newly prosperous traveling public.

Route 66 inspired a catchy song by Bobby Troup, “(Get Your Kicks on) Route 66,” and starred in TV shows and movies.

With its drive-ins, motels and tourist traps, Route 66 got Americans to hit the road — then bombarded them with marketing gimmicks to get them to pull over, such as giant concrete wigwams or fiberglass dinosaurs.

Snakes were a recurring theme. An exhibit now at the New Mexico Museum of Art in Santa Fe, “Roadside Attractions” (through Oct. 4), includes a black-and-white photo by Steve Fitch showing a tall, skinny man holding a snake in front of a large sign for the Regal Reptile Ranch & Menagerie, on Highway 66 in Sayre, Okla.

Image

“Roadside Attractions,” an exhibit at the New Mexico Museum of Art in Santa Fe, includes this photo by Steve Fitch, taken on Highway 66 in Sayre, Okla.Credit...Steve Fitch; Photo by Blair Clark

While not all the 20-plus photos in the show are from Route 66 in particular, they evoke the feeling of it — “the extraordinary experience of driving and being on the road and encountering these oddities,” said Katherine Ware, the museum’s former curator of photography, who put together the exhibition.

In their 2025 book “Route 66: The First 100 Years,” Jim Ross and Shellee Graham called the highway itself “a linear museum stretched across eight states.”

Westward Bound

The road’s westward route started in downtown Chicago, dropped down through Illinois and Missouri and cut through a corner of Kansas before crossing Oklahoma, the Texas Panhandle, New Mexico, Arizona and Southern California.

But geographic details of the route have evolved. In fact, in February, Chicago declared that the route’s symbolic starting point would now be at Navy Pier. (The route’s longtime official starting point is across the street from the Art Institute of Chicago.)

Springfield, Mo., touts itself as the birthplace of Route 66 because a telegram sent from there on April 30, 1926, settled a protracted bureaucratic dispute over the road’s designation, establishing that it would carry the number 66.

Image

The upcoming 100th anniversary kickoff in Springfield, Mo., will include an opening concert and a parade of classic vehicles. Credit...Chris Akins, City of Springfield, Missouri

The actual anniversary of the highway is Nov. 11, the date when the American Association of State Highway Officials formally adopted the U.S. numbered highway system as a whole.

The upcoming kickoff in Springfield will include an opening concert and a parade of classic vehicles. Visitors can drop in to the local History Museum on the Square, which has a dedicated Route 66 gallery. From June to September, the museum will also have a separate exhibit, “The Route 66 Series: 1926-2026,” examining the ways the road shaped aspects of local culture, including through cars, music, fashion and the economy.

A local project is also in the works to tell a more somber side of the Route 66 story, with an exhibit tentatively titled “The mOTHER Road: The Stories of African Americans along Historic Route 66,” expected to open by June at the Benton Avenue A.M.E. Church in Springfield.

Lyle Q. Foster, an associate professor of sociology at Missouri State University who is leading the effort, said that because of racial segregation, African Americans relied on the Green Book, a travel guide that listed safe places where they could eat, sleep or fill their gas tanks. In some “sundown towns,” he said, people of color were not even supposed to show their presence after dark.

“It certainly was a very challenging road to travel,” Dr. Foster said. “It wasn’t always getting your kicks and fix on Route 66 for people who were not white.”Councilwoman Monica Horton of Springfield, who represents the historically Black part of the city and is helping put together the exhibit, said it was important for the community to honor the legacy and resilience of the Green Book businesses and “to see ourselves in these stories.”

Multiple Events

The kickoff is one of dozens of events approved by the Route 66 Centennial Commission, established by statute in 2020 to identify appropriate activities to honor “the Mother Road of the United States.”

The commissioners, who are unpaid presidential appointees, have focused on larger-scale celebrations, but “virtually all of the small communities along the route will have some type of an event,” Rod Reid, the commission chair, said.

Then there are the small museums devoted to the lore of Route 66, or the cars that traveled along it.

In McLean, Texas, the Devil’s Rope Museum mainly focuses on barbed wire, but its Route 66 room is what usually draws people in from the freeway, said the museum’s president, Michael Oldham. There’s a connection between the two subjects, he said: Without the use of barbed wire along Route 66, “all you’d do is dodge cattle.”

In Illinois, the Joliet Area Historical Museum welcomes about 50,000 people annually to its Route 66 Visitors Center, about a quarter of them from outside the United States, according to Greg Peerbolte, the museum’s chief executive.

Many of the international visitors, he said, expressed the same sentiment about the highway: “This is the real America.”

Americans, Peerbolte added, tend to be drawn by nostalgia for their own experiences on the road.

The Illinois State Museum created a Route 66 Oral History Project, interviewing 100 people who either knew the highway in its heyday or have worked to preserve it since. On Memorial Day weekend the museum plans to open a companion exhibition of physical objects, “Miles of Memories: Stories of Route 66.”

Even today, people who travel the road have a different experience than they would if they stuck to interstates and hotel and restaurant chains, said Bill Thomas, the chair of the Route 66 Road Ahead Partnership, which works to promote and revitalize the road.

“The interstate highway system cannot deliver memories the way Route 66 can,” he said.

Myth and Nostalgia

David Davis, chief curatorial officer at the National Cowboy & Western Heritage Museum in Oklahoma City, said the appeal of Route 66 was rooted partly in nostalgia for the 1950s, but also in the mystique of the American West, which he described as “so interconnected to American identity itself.”

Image

“Route 66: From Trails to Truck Stops,” at the National Cowboy & Western Heritage Museum in Oklahoma City, includes a large neon sign with a cowboy on horseback, advertising the now-defunct Western Trail Motel. Credit...National Cowboy & Western Heritage Museum

“It is just a part and piece of that American cultural puzzle,” he said in an interview. The museum’s exhibition “Route 66: From Trails to Truck Stops” (through May 3) includes a large neon sign with a cowboy on horseback, advertising the now-defunct Western Trail Motel, as well as a 1948 blue Chrysler Windsor.

Route 66 is part of the museum’s own heritage, Davis said. He explained that, in the 1950s, when cities were competing to host the museum, Oklahoma City won out partly because of its proposed site on Persimmon Hill, overlooking the famed highway.

At the Museum of Northern Arizona in Flagstaff, the exhibition “Wagon Road to Mother Road” (through Jan. 31) challenges some of the nostalgia associated with Route 66 by examining some of the objects that were discarded along the way.

David E. Purcell, the museum’s supervisory archaeologist and historian, curated the exhibition, drawing in part on an archaeological survey he worked on in collaboration with the Petrified Forest National Park. Along a six-mile stretch of the old highway, the team designated sample areas and documented more than 11,000 artifacts, which Purcell estimated probably amounted to just 10 to 15 percent of what was there.

About one-third of the cans and bottles they found had once contained alcoholic beverages, he said. They also found part of a cocktail shaker, “church keys” for opening cans of beer and evidence of car accidents.

Flagstaff benefits from Route 66 tourism today, but people forget that the old road was narrow, nerve-racking and dangerous, Purcell said.

“By the time it disappeared, there was no nostalgia for it,” he said. “It wasn’t until literally they were pulling down the signs that some of that nostalgia began.”

NEW YORK TIMES 

Peter Max, Trippy Artist of Counterculture Era and Beyond, Dies at 88

 

 

 

With work that breathed the spirit of the 1960s, he became fabulously wealthy from licensing his paintings and posters in all their euphoric Day-Glo glory.

William Grimes 

Peter Max, whose trippy posters of Day-Glo flowers, butterflies, rainbows, stars and planets sold by the millions and expressed the blissed-out exuberance of the countercultural 1960s, died on Monday in Manhattan. He was 88.

The death, at a hospital, was confirmed by his son, Adam Max. He had long suffered from dementia.

Mr. Max, a vegetarian and a teetotaler who expanded his consciousness through yoga, not drugs, was one of the most visible and successful artists of his time, selling posters in the millions.

His simple graphic images, brightly colored and infectiously happy, lent themselves to a long list of licensed commercial products, from sheets and glassware and shower curtains to airplanes and, on one occasion, a stamp for the Postal Service.

Incorporating Art Nouveau, Op Art and psychedelic motifs, Mr. Max’s designs offered instant visual gratification and a cheery endorsement of the generational politics of love, joy and universal consciousness. His “panopticon” images repeated visual elements in tight formation, producing a kaleidoscope effect that recalled Eastern mandalas.

Image

“Cosmic Runner.”Credit...Peter Max, via Park West Gallery

His “cosmic jumper,” a young man in bell bottoms traversing the globe in giant strides, became an emblem of modern youth, eager to explore the far reaches of inner and outer space. Barefoot and surrounded by puffy clouds and chubby birds, the jumper appeared on the 10-cent stamp commissioned for Expo ’74, the conservation-themed World’s Fair in Seattle.

In “The Universe of Peter Max,” a memoir published in 2013, he wrote: “My cosmic art subjects were always euphoric: Himalayan sages levitating under a shower of flowers beside snowcapped mountain peaks; faces and flowers exploding into a symmetrical cosmic butterfly; mirrored lakes and waters with a cascade of stars and planets; flowers in perpetual states of organic growth, begetting blossoms. My inspiration came from the new youth movement in America and in return spoke to it.”

Peter Max Finkelstein was born on Oct. 19, 1937, in Berlin, where his father, Jakob, was in the import-export business and his mother, Salla (Zeisel) Finkelstein, had been a fashion designer before her marriage. With the rise of Nazism, the family moved to Shanghai, where Jakob set up as a pearl merchant. Peter attended an English-language school and spent much of his spare time talking to monks at the nearby Buddhist monastery.

As the Maoist revolution gathered strength, the Finkelsteins moved again, this time to Haifa, in Israel, where Peter learned Hebrew and, infatuated with the stars and planets, took special courses in astronomy at the Technion, the Israel Institute of Technology.

Image

Mr. Max in 1973 in his studio. Through the half-dozen businesses bearing his name, he entered into licensing agreements with more than 70 companies.Credit...Dave Pickoff/Associated Press

In 1953, the family emigrated to New York, settling in Bensonhurst, Brooklyn, where Peter attended Lafayette High School. He had taken private art lessons in Israel, and, at the urging of a classmate, signed up for summer classes at the Art Students League in Manhattan after high-school graduation. Abandoning his rather vague plan to pursue astronomy as a career, he threw himself into art.

Initially he painted as a realist, then dabbled in abstraction and finally concentrated on commercial art. Riverside Records gave him his first assignment, to design the LP sleeve for “The Blues Piano Artistry of Meade Lux Lewis,” released in 1962. His moody, vaporous portrait of Mr. Lewis at the piano won a gold medal at the annual exhibition of the Society of Illustrators.

With Tom Daly, he formed a design company that quickly found favor with advertising agencies and book publishers. Within two years, Daly & Max Studio had collected more than 60 awards for its book jackets, album covers and other work, but Mr. Max, feeling constrained, left in 1965. For two years, he lived in virtual seclusion in his duplex apartment on Riverside Drive, turning out thousands of drawing and ideas.

Image

Mr. Max in 2012. He designed the logo for MTV in the 1980s and the official poster for the World Cup in 1994.Credit...Simon Russell/Getty Images

In 1967, he applied some of his new concepts to the Tin Lizzie in midtown, a theme restaurant that he decorated with paisley wallpaper, photographic collages of Charlie Chaplin film scenes and an actual Model T Ford.

His advertising posters for local shops began to attract special attention. Young customers called up the job printer, Security Printing Company, asking for a poster to hang in their rooms. Denton Stein, the company’s owner, went into partnership with Mr. Max to create Peter Max Posters, which sold 25 of the artist’s designs, in millions of copies, to book stores, head shops and department stores.

As his popularity grew, Mr. Max began doing commercial posters for the American Cancer Society, the Peace Corps and the United Nations. The 3,000 posters and 40,000 fliers he printed for a gathering in Central Park on Easter 1967 inadvertently gave the event its name. The words “Be In,” in large letters, followed by “Central Park on March 31,” were read as a noun, not a verb in the imperative mood.

In 1968, a transit advertising company commissioned Mr. Max to design posters for buses, streetcars and subways in 10 major cities. “I am bringing art to the people where the people are,” he told Time magazine. “Now they don’t have to go to galleries and see paintings on a wall.”

General Electric approached him to design a clock. He followed up with a line of linens for J.P. Stevens and shirts for Van Heusen. Through the half dozen businesses bearing his name, he entered into licensing agreements with more than 70 companies.

“You provide the cash, the production and the selling,” he told Business Week in 1968. “I’ll furnish the ideas and the designs — and we’ll split the profits 50-50.” When Life magazine put Mr. Max on its cover in 1969, the headline read “Peter Max: Portrait of the Artist as a Very Rich Man.”

Image

The cover of Life magazine in September 1969.Credit...Life Magazine, via Park West Gallery

Mr. Max was everywhere. He was not, however, involved with the 1969 Beatles film “Yellow Submarine,” often assumed to be his work, in part because he often took credit for it. The film’s art director was Heinz Edelmann, sometimes called the German Peter Max.

By the 1970s, Mr. Max was in a rut. He shut down his studio and again went into seclusion. “I stopped cold,” he told People magazine in 1976. “I was repeating myself creatively. I had filled myself with being everywhere, doing everything. I had had it.”

Image

Mr. Max working on a painting of the Statue of Liberty in New York Harbor as art of the Fourth of July celebration in 1987.Credit...David Bookstaver/Associated Press

Gradually, he re-emerged. For the American Bicentennial, he accepted a commission from the Swedish company ASEA to produce the book “Peter Max Paints America,” a collection of paintings and drawings honoring each of the 50 states.

That same year, he produced a series of bilingual “Welcome to the United States” signs posted at more than 200 crossings along the Canadian and Mexican borders. Initially rejected by the Customs Service as implicitly endorsing the ’60s drug culture, they were finally put up in 1978.

He designed the logo for MTV in the 1980s and the official poster for soccer’s World Cup in 1994. In 1999 he decorated a Boeing 777 for Continental Airlines to commemorate NYC 2000, the official millennium celebration in New York. In 2015, the Rock & Roll Hall of Fame commissioned him to do a poster for its annual induction ceremony.

Image

“LOVE.”Credit...Peter Max, via Park West Gallery

By then, he was already struggling with dementia. Yet the Peter Max brand flourished, boosted by the artist’s relationship with Park West Galleries, a Michigan-based company that sold Mr. Max’s work at cruise-ship auctions. Sales on a single voyage could reach $2 million, and in 2018, Mr. Max’s studio, ALP Inc. — named for its three owners, Peter Max and his two children, Adam and Libra — recorded a profit of more than $30 million.

The artist’s production methods became the subject of widespread speculation. Friends reported that Mr. Max could no longer paint and merely signed his work, which was made, in prodigious quantities, by a small army of assistants.

Image

Mr. Max painted a series of portraits of firefighters who died in the line of duty, including some who were killed in the 9/11 attacks.Credit...Getty Images/Getty Images

They were overseen by Adam Max, who was ousted from his father’s business in 2019 by his daughter, Libra Max, and the artist’s court-appointed guardian, who were concerned that Mr. Max was being exploited. Ms. Max took over her father’s studio, fired many of his assistants and business associates, and set about restoring his reputation through galley and museum shows that showcased his earlier work.

Mr. Max’s first marriage, to Elizabeth Nance, with whom he had the two children, ended in divorce. In 1997, he married Mary Balkin, who was 30 years his junior and whom he randomly met on a Manhattan sidewalk a year earlier. “Hi, I’m Peter Max, and I’ve been painting your profile my entire life,” he reportedly said upon first seeing her.

She died by apparent suicide in 2019, amid her husband’s worsening dementia and an extended legal dispute with her stepson over the handling of Peter Max’s art and wealth.

Mr. Max is survived by Adam and Libra Max.

For his legions of fans, Mr. Max would always remain an avatar of the 1960s counterculture in its happiest, most innocent phase. Whatever the year, and whatever the occasion, his work always breathed the spirit of the 1960s. When the De Young Museum in San Francisco organized an exhibition in 2007, the title said it all: “Peter Max and the Summer of Love.”

“Just as The Beatles had the music, I had the images,” he told Crain’s New York Business in 2007. “Kids would buy the posters and look at them as they listened to the music. It was like their MTV.”

NEW YORK TIMES