Two attacks on Sam Altman in 48 hours. AI has a consent problem no policy paper can fix.

On April 6, Sam Altman published a 13-page document called “Industrial Policy for the Intelligence Age.” It proposed a public wealth fund, a robot tax, and a four-day workweek. It read like a campaign platform for a candidate running for president of the future.

Four days later, at 3:45 in the morning, a 20-year-old from Texas named Daniel Alejandro Moreno-Gama threw a Molotov cocktail at Altman’s San Francisco home. The bottle hit the metal gate, caught fire, and was put out by security guards. Nobody was hurt. Police arrested Moreno-Gama that afternoon. He had been writing on Substack about AI causing human extinction. He had posted in a PauseAI Discord server in December: “We are close to midnight, it’s time to actually act.”

Two days after that, on Saturday morning, a car drove past the same house and someone fired a gun at it. Two more people arrested. Amanda Tom, 25, and Muhamad Hussein, 23, charged with negligent discharge of a firearm.

Two attacks in 48 hours. On the home of the most visible person in AI.

I am not writing this to litigate the attacks themselves. Violence against anyone is wrong, and I want to be clear about that before going further. PauseAI put out a statement saying violence is antithetical to everything they stand for. The attacker acted alone.

But I keep thinking about the timing. A policy paper on Monday. A firebomb on Friday. A shooting on Sunday. And in between, a Quinnipiac poll showing that 80% of Americans are concerned about AI.

That is the story I want to talk about. Not the violence. The gap.

The numbers that should worry every AI builder

The Quinnipiac poll came out on March 30. I have read a lot of AI surveys over the past year. This one hit different.

80% of Americans are concerned about AI. Not 50%. Not a slim majority. Eight out of ten people.

55% believe AI will do more harm than good in their daily lives. That number is up 11 points from April 2025. The direction is not ambiguous. It is getting worse, not better.

70% believe AI advances will reduce job opportunities. That was 56% a year ago. A 14-point jump in 12 months.

Gen Z, the generation that grew up with this stuff, is the most pessimistic. 81% of them think AI will reduce jobs. Not boomers. Not people who struggle to use their phones. The generation that actually understands the technology is the most afraid of it.

And here is the number that I think matters most: only 21% of Americans trust AI-generated information most or almost all of the time. Meanwhile, 51% are using AI for research. People are using tools they do not trust. That is not adoption. That is dependency without consent.

What Altman actually said

After the first attack, Altman wrote a blog post. He said: “The fear and anxiety about AI is justified.”

He also wrote: “We are in the process of witnessing the largest change to society in a long time, and perhaps ever.”

I want to sit with those two sentences together. The CEO of the company most responsible for this change is saying both that the anxiety is justified and that the change might be the biggest in human history. He is acknowledging the scale and the fear simultaneously. And his response to both is a 13-page policy paper that proposes things Congress has already rejected.

The public wealth fund? Proposed four times. Never got committee hearings. The 32-hour workweek? Never made it to a vote. The robot tax? Never got a bill number. Soribel Feliz, a former Senate AI adviser, said there is a gap “between naming solutions and building real mechanisms.” Anton Leicht at the Carnegie Endowment called the whole thing “comms work to provide cover for regulatory nihilism.”

I do not think Altman is lying. I think he is stuck in the same place every AI leader is stuck. He knows the anxiety is real. He knows the technology is moving faster than any institution can absorb. And the only tools he has are more technology and more policy papers. Neither one addresses the actual problem.

The actual problem is consent

When I say consent, I do not mean terms of service. I mean something closer to what political scientists call “social license to operate.” The idea that a company or industry can only function when the broader public accepts that it should exist and operate the way it does.

The oil industry had this problem. Tobacco had this problem. Social media has this problem right now. And AI is developing it faster than any of them.

The difference is speed. It took decades for public opinion to turn against tobacco. Social media had about 15 years before the backlash started. AI has been mainstream for roughly three years, and 80% of the country is already concerned.

I covered Anthropic passing OpenAI in revenue last week. One of the reasons Anthropic is winning enterprise deals is that they maintained safety guardrails the Pentagon wanted removed. Over 100 enterprise customers called Anthropic after the Pentagon blacklisting to check in, not to cancel. The companies paying millions for AI want to know their vendor will not embarrass them. That is a version of the consent problem playing out at the B2B level.

But at the consumer and citizen level, nobody is asking for permission. Nobody asked the residents of Madison, Wisconsin whether they wanted a data center drawing power from their grid. Nobody asked the artists whose work trained the models. Nobody asked the 70% of Americans who think AI will take their jobs whether they wanted this to happen at this speed.

The protest movement is not fringe anymore

I want to lay out what is actually happening on the ground, because the coverage tends to focus on the most dramatic incidents and miss the pattern.

In February, London’s King’s Cross tech hub saw what PauseAI described as the largest anti-AI protest yet. The group has been growing on what one observer called “a somewhat exponential path, matching the progress of AI itself.”

Local data center bans are spreading across the United States. Madison, Wisconsin passed one. New Orleans passed one. Communities across Georgia and Michigan have passed similar restrictions. No statewide moratoriums have succeeded yet, but the local bans are multiplying.

Nearly 1,000 workers from OpenAI and Google signed an open letter demanding their companies refuse Pentagon contracts. These are not outsiders. These are people building the technology.

The attacks on Altman’s home are the most extreme manifestation of something that has been building for months. Data center opposition. Military AI opposition. Job displacement fear. Existential risk belief. These are four different movements with four different motivations, and they are all growing at the same time.

Steve Bannon, who is not someone typically aligned with AI safety activists, called for a pause on superintelligence development. When your backlash coalition includes both PauseAI volunteers in London and Steve Bannon, the ground has shifted underneath you in ways that a policy paper cannot address.

The “Butlerian Jihadist” and the gap between online and offline

I want to talk about the attacker for a moment, not to sympathize with him, but because his path tells us something about the information environment around AI.

Daniel Moreno-Gama used the screenname “Butlerian Jihadist,” a reference to the Dune novels where humanity destroys all thinking machines. He ran a Substack where he wrote about AI as an extinction risk. He described tech leaders as “sociopathic and reckless.” He specifically called Altman “a pathological liar gambling with humanity’s future.”

In December, he posted in the PauseAI Discord: “We are close to midnight, it’s time to actually act.” A moderator responded immediately, warning that advocating violence was grounds for an instant ban.

PauseAI is a legitimate advocacy organization. They do not promote violence. But they exist in an information ecosystem where mainstream AI researchers, credentialed people with PhDs and published papers, say things like “there is a meaningful probability AI causes human extinction.” Geoffrey Hinton won the Nobel Prize and says he sometimes regrets his life’s work. Yoshua Bengio, another Turing Award winner, has called for international oversight because the risk is real.

When credentialed, respected scientists say the technology might end civilization, some fraction of people will take that literally and act on it. That is not PauseAI’s fault. It is not Hinton’s fault. But it is a predictable consequence of the information environment, and the AI industry has done almost nothing to address it.

The standard response from AI companies is: we take safety seriously. But “we take safety seriously” is a statement about your internal process. It is not an answer to the person who just watched a YouTube video where a Nobel laureate says AI might kill everyone. The gap between “we take safety seriously” and “here is why you should not be terrified” is enormous, and nobody is filling it.

Why the New Deal paper makes things worse, not better

OpenAI’s policy document is fascinating as a political artifact. It reads, as one critic noted, like something “distinctly Sanders-coded.” Public wealth fund. Robot tax. Higher capital gains taxes. Four-day workweek. If you did not know who published it, you might guess it came from a progressive think tank.

But it came from a company that completed a for-profit restructuring last year, cultivated close ties to the Trump administration, and just raised the largest venture round in history at $122 billion. The contrast between the populist proposals and the corporate reality is hard to miss.

More importantly, the document asks government to solve problems that AI companies created. Every proposal is a request for someone else to build the safety net: Congress should create the wealth fund. Congress should pass the robot tax. Congress should incentivize the shorter workweek. OpenAI will continue building superintelligence and hopes government catches up.

Lucia Velasco, a former U.N. AI policy head, pointed out that OpenAI is shaping “the environment in which OpenAI operates with significant freedom.” That is a diplomatic way of saying: the company is writing the rules for its own regulation while accelerating the technology that makes regulation necessary.

I wrote about TSMC’s $35.7 billion quarter yesterday and the lesson was about where AI profits actually concentrate. The lesson here is similar. The people capturing the value from AI are not the same people absorbing the disruption. And a policy paper from the value-capturer asking the disrupted to be patient is not going to close the trust gap. If anything, it widens it.

What the AI industry is getting wrong about trust

There is a framework I keep coming back to. Companies think about trust as a communications problem. Put out a blog post. Publish a safety paper. Hire a head of public policy. Get Sam to do a podcast.

But trust is not a communications problem. Trust is a power problem.

People do not distrust AI because they lack information. 51% of them are using it for research. They have plenty of information. They distrust AI because they feel powerless in relation to it. They cannot opt out. They cannot slow it down. They cannot negotiate the terms. The technology is changing their jobs, their kids’ schools, their information environment, their creative industries, and their energy grid, and they have no mechanism to say “not this fast” or “not like this.”

That is what consent means. Not “did you read the terms of service” but “did anyone ask whether you wanted this?”

The Quinnipiac poll captures this perfectly. 51% use AI. 55% think it does more harm than good. 21% trust it. People are using something they believe is harmful and do not trust. That is not an adoption curve. That is the psychology of a population that feels it has no choice.

Altman himself acknowledged this. “The fear and anxiety about AI is justified.” But acknowledging fear while continuing to accelerate is the opposite of building trust. It is the equivalent of a doctor saying “your concerns about this drug are valid” while increasing the dosage.

What builders need to understand about social license

I write this blog for founders. So here is the part that matters to anyone building an AI company right now.

Social license to operate is not something you earn once. It is something you maintain continuously. And the AI industry as a whole is losing it.

If you are building an AI product, the political environment around that product is changing faster than your product roadmap. Local data center bans can affect your infrastructure. Public opinion on AI jobs can affect your hiring brand. Pentagon controversies can affect whether enterprise customers want to be associated with AI companies. A single violent incident can shift public sentiment in ways that take years to repair.

You might think this only applies to OpenAI or Anthropic or Google. It does not. When 80% of the country is concerned about AI, they are concerned about AI, not about specific companies. The backlash is categorical. Your startup that automates dental office scheduling is in the same category as the superintelligence lab, in the public’s mind.

Three things I think every AI founder should be doing right now.

First, be specific about who your product helps and how. “AI-powered” is not a value proposition anymore. It is a liability flag. If your marketing leads with the technology rather than the outcome, you are inviting the skepticism that 80% of the country already feels.

Second, make opt-out easy and visible. Give users control. Let them see what the AI is doing. Let them turn it off. The companies that will survive the backlash are the ones that gave people a genuine choice, not the ones that embedded AI so deeply it could not be separated.

Third, think about the people who are not your users. Your product may help your customers. But does it displace your customers’ employees? Does it use your customers’ data without their knowledge? Does it draw power from your customers’ communities? The consent problem is not between you and your user. It is between the AI industry and everyone else.

The timeline that matters

Here is what happened in one week.

April 6: OpenAI publishes its New Deal for AI.

April 8: I wrote about OpenAI exploring ads in ChatGPT. The company that just proposed a robot tax to protect workers is also figuring out how to monetize the attention of 400 million users.

April 10: Molotov cocktail at Altman’s home. The attacker believed AI would end humanity. He had been radicalized online over months, consuming a steady diet of legitimate scientific concern filtered through doomer communities.

April 10: I wrote about Anthropic passing OpenAI in revenue while the Pentagon tried to destroy it for keeping safety guardrails. Enterprise customers rewarded constraints. The Pentagon punished them. Both facts were connected.

April 11: I covered Claude Mythos finding thousands of zero-day vulnerabilities with a 72.4% exploit rate. The Treasury Secretary called an emergency meeting. The same company the Pentagon blacklisted built an AI so powerful it scared the financial system.

April 12: A second attack on Altman’s home. Gun fired from a passing car.

April 12: I wrote about TSMC making $35.7 billion in a single quarter. The company making the chips that make AI possible is more profitable than almost every AI company combined.

One week. A policy paper, two attacks, a cybersecurity emergency, a Pentagon controversy, an enterprise trust story, a $35.7 billion earnings report, and a poll showing 80% of the country is afraid.

That is the speed at which this is moving. And the gap between that speed and the public’s ability to process, consent, and adapt is the defining challenge of AI in 2026. Not capabilities. Not funding. Not regulation. Consent.

Where this goes

I do not have a clean answer. I do not think anyone does.

But I think the path forward involves AI companies doing something they have not been willing to do: slow down voluntarily in specific areas where public consent has not been established. Not because regulators forced them. Not because it is good PR. Because the alternative is a backlash that makes regulation inevitable and probably worse than whatever self-imposed constraints they would have chosen.

The oil industry learned this lesson. Tobacco learned it. Social media is learning it right now, badly and too late. AI has a chance to learn it faster, but only if the people building it stop treating public concern as a communications problem and start treating it as a power problem.

80% of Americans are concerned. 55% think AI does more harm than good. Two people attacked Sam Altman’s home in 48 hours. And the industry’s response is a 13-page document proposing a robot tax that Congress has already rejected.

The gap is the story. And it is getting wider.

Frequently asked questions

What happened at Sam Altman’s house in April 2026?

On April 10, 2026, Daniel Alejandro Moreno-Gama, a 20-year-old from Texas, threw a Molotov cocktail at Sam Altman’s San Francisco home at approximately 3:45 a.m. The device hit a perimeter gate and was extinguished by security. Moreno-Gama was arrested that afternoon and charged with attempted murder, arson, and possession of an incendiary device. He later appeared outside OpenAI’s office and threatened to “burn down the building.” Two days later on April 12, a second attack occurred when a car drove past and someone fired a gun at the property. Amanda Tom, 25, and Muhamad Hussein, 23, were arrested and charged with negligent discharge of a firearm. No one was injured in either attack.

Why did someone attack Sam Altman’s home?

The first attacker was motivated by fears that AI development would lead to human extinction. Using the screenname “Butlerian Jihadist” (a Dune reference about destroying thinking machines), Moreno-Gama ran a Substack about existential AI threats, called tech leaders “sociopathic and reckless,” and specifically labeled Altman “a pathological liar gambling with humanity’s future.” In December 2025, he posted “We are close to midnight, it’s time to actually act” in the PauseAI Discord server. PauseAI immediately distanced itself from the attack, saying violence is antithetical to their mission.

What is OpenAI’s New Deal for AI?

On April 6, 2026, OpenAI published “Industrial Policy for the Intelligence Age,” a 13-page policy document proposing a public wealth fund giving every American a stake in AI-driven growth, a robot tax, a four-day 32-hour workweek at full pay, higher capital gains and corporate income taxes, and framing AI access as a fundamental right. Critics noted that every proposal corresponds to legislation that previously failed in Congress, with the 32-hour workweek never reaching a vote, the wealth tax proposed four times without committee hearings, and the robot tax never getting a bill number.

How do Americans feel about AI in 2026?

According to a Quinnipiac poll from March 30, 2026: 80% are concerned about AI (38% very concerned, 42% somewhat concerned). 55% believe AI does more harm than good, up 11 points from 2025. 70% believe AI will reduce jobs, up from 56% a year earlier. Gen Z is the most pessimistic group, with 81% expecting job losses. Only 21% trust AI-generated information, despite 51% using it for research. 64% believe AI is more harmful than helpful in education.

Is there a growing anti-AI protest movement?

Yes. PauseAI has been growing rapidly. London saw its largest anti-AI protest in February 2026 at the King’s Cross tech hub. Local data center bans are spreading in Madison, New Orleans, Georgia, and Michigan. Nearly 1,000 OpenAI and Google workers signed a letter opposing Pentagon contracts. The movement spans multiple concerns: existential safety, job displacement, military applications, environmental impact of data centers, and creative industry disruption. Steve Bannon called for a superintelligence pause, indicating the backlash crosses traditional political lines.

What should AI founders learn from the attacks on Sam Altman?

Social license to operate is a prerequisite for sustainable growth. When 80% of the country is concerned about AI, the backlash is categorical and affects every AI company, not just the largest ones. Founders should lead marketing with outcomes rather than “AI-powered” branding, make opt-out easy and visible, and consider the impact on people who are not their direct users, including displaced workers, affected communities, and data subjects. The gap between AI building speed and public consent speed is the defining challenge of 2026, and it is widening. Companies that treat public concern as a power problem rather than a communications problem will be better positioned when regulation inevitably arrives.