Someone firebombed Sam Altman's house, then told security he'd burn down OpenAI with everyone inside.

The Summary

The Signal

The AI safety debate just left the internet. For years, the argument about whether advanced AI poses an existential threat has been confined to Twitter threads, academic papers, and awkward conference panels. Now someone turned it into a firebomb.

Moreno-Gama's manifesto called out the disconnect between Altman's public warnings about AI dangers and OpenAI's relentless push to build more powerful models. He's not wrong about the contradiction. Altman has spent years warning about AI risks while simultaneously racing to achieve AGI. That tension has fed a narrative that AI labs are playing Russian roulette with humanity's future for profit.

"When the CEO warns about extinction while shipping new models every quarter, someone was bound to take the warnings literally."

The timing matters. AI coding tools are accelerating development cycles. Model releases are happening faster. The gap between GPT-4 and GPT-5 capabilities will likely be larger than the gap between GPT-3 and GPT-4. The perceived rate of change is fueling anxiety, even among people who use these tools daily.

But here's what the violence obscures: the actual risk calculation. The doomer argument rests on a chain of assumptions:

  • Current scaling trends continue indefinitely
  • Intelligence implies goal-seeking behavior that conflicts with human values
  • We won't figure out alignment before we hit the danger zone

None of these are certainties. But uncertainty cuts both ways. When you tell people they might be building the thing that ends humanity, and then you keep building it anyway, you create the conditions for radicalization.

The Fast Company piece notes that some interpret Altman's danger warnings as humble-brags about OpenAI's capabilities. That's probably half right. The warnings serve multiple purposes: they signal that OpenAI is working on genuinely powerful technology, they position Altman as a responsible actor thinking about long-term risks, and they provide cover for regulatory capture by making AI sound so dangerous that only the biggest players should be trusted to build it.

What's new is that someone took the warnings at face value and decided direct action was the only rational response. If you genuinely believe AGI will kill everyone and the people building it won't stop, violence starts to look like self-defense. That's the logic of eco-terrorism, anti-abortion extremism, and now, apparently, AI doomerism.

"The manifesto didn't just target Altman. It called for killing other AI CEOs and their investors."

This won't be the last incident. The doomer community is small but growing, fueled by AI researchers who've left the field over safety concerns and amplified by effective altruist circles that treat existential risk as the only cause worth caring about. Most will never turn violent. But the same dynamics that radicalized environmental activists and anti-globalization protesters are now present in AI discourse: a perceived existential threat, powerful actors who won't listen, and a sense that time is running out.

The Implication

AI companies now have to factor physical security into their risk models. Expect armed guards, hardened facilities, and tighter operational security around key researchers and executives. That's a tax on innovation, and it will slow things down, but not in the way safety advocates hoped.

For builders in the agent economy, this changes nothing technically but everything politically. The narrative around AI is shifting from "this will change everything" to "this might kill everyone." That makes it harder to raise money, hire talent, and ship products without constant justification. If you're building agents that automate real work, be ready to explain why your specific application doesn't accelerate the path to doom. It's exhausting, but it's the new cost of doing business.

Sources

Fast Company Tech