This is not just another tech news story. This is the first real explosion of the AI era.
On April 11, 2026, Beijing time, OpenAI CEO Sam Altman posted on his blog: the previous day, in the early hours of the morning local time, a 20-year-old man threw a Molotov cocktail at his residence in San Francisco. The device bounced off and caused no injuries. Police quickly arrested the suspect.
There is currently no direct evidence linking the attack to AI controversies — yet Altman himself explicitly connected it to the social anxiety triggered by AI.
Why? Because he understands: this fire did not burn his house’s exterior wall; it burned through the long collective slumber of the entire AI elite circle.
I. A Louder Alarm Than Any AI Debate
For the past few years, debates about AI have never ceased: accelerationism vs. decelerationism, Musk vs. Altman, OpenAI’s internal struggles, IPO plans, effective altruism… But all of these remained confined to elite circles — academic papers, tweets, and court filings. Ordinary people were merely distant spectators.
The Molotov cocktail shattered that illusion.
It was not theory, not a petition, not an open letter. It was a 20-year-old man expressing, in an extreme, wrong, yet profoundly real way, a despair that had been suppressed to the breaking point:
“I cannot influence the direction of AI through any normal channel, so I resorted to the most primitive form of violence.”
This incident marks a watershed not because it “might” be related to AI, but because the social anxiety caused by AI has, for the first time, transformed from an abstract concept into a real physical flame.
Previously, when we said “AI may trigger social unrest,” it was a prediction. Now, a Molotov cocktail has landed on a Silicon Valley lawn.
From this point on, any discussion about AI’s societal impact that continues to pretend it is merely a topic for conference rooms or Twitter is self-deception.
II. Violence Has No Excuse, But the Fear Is Not an Illusion
Throwing a Molotov cocktail is a serious assault on the rule of law and has no justification whatsoever. Altman’s blog post, in which he shared family photos, acknowledged his own shortcomings, and called for reducing confrontation, demonstrated commendable restraint.
However, if we only condemn the violence while avoiding the two core questions — “Why him? Why now?” — we are evading reality.
Sam Altman is the helmsman of the world’s most prominent AI company, the frontrunner in the AGI race, and the iconic figure of “accelerationism.” To countless ordinary people, he is the man who is deciding humanity’s future on their behalf — even if he may not see himself that way.
When someone is viewed as the bearer of the “One Ring,” while his company shifts from nonprofit to full-profit mode, prepares for an IPO, and battles former partners in court, the extreme hostility he faces — though absolutely unacceptable — does not arise from nowhere.
The Molotov cocktail is an ugly form of expression. But the root cause is real: too many people feel they have completely lost their voice in the AI era, and the speed of technological change has far exceeded the limits of societal psychological endurance.
III. The Three-Layer Structure of AI Fear: Increasingly Real and Increasingly Terrifying
First Layer: The Collapse of Jobs and Dignity
Cars replaced horse-drawn carriages, but AI is replacing human cognitive labor itself.
Just days ago — on April 7, 2026 — Anthropic released Claude Mythos Preview, internally described as the “strongest Claude in history.”
This model has made a stunning leap in code understanding, complex reasoning, and vulnerability discovery. It can autonomously identify thousands of high-risk zero-day vulnerabilities, including:
- A 27-year-old vulnerability in OpenBSD
- A 16-year-old vulnerability in FFmpeg
- Vulnerabilities in the Linux kernel and even chain multiple vulnerabilities together to achieve privilege escalation.
More alarming is that the model is currently not open to the public. Anthropic has stated explicitly that it is “too powerful and too dangerous,” fearing malicious use for cyberattacks. It is therefore restricted to a small number of partners for defensive security research — the Project Glasswing initiative.
What is truly terrifying: if it were fully released, entire industries such as security auditing, penetration testing, and code review could collapse overnight. Millions of jobs that rely on deep professional expertise would be replaced by AI in an extremely short time.
This uncertainty — “a respectable profession today could disappear collectively tomorrow” — is creating genuine, persistent panic in countless hearts.
Second Layer: Extreme Concentration of Power
Critical decisions about AI development are made by a handful of laboratories and tech giants. Ordinary people cannot even find an effective “disagree” button.
Third Layer: Intergenerational Injustice
Today’s AI trajectory is dominated by elites and capital, but the full consequences will be borne by the next generation and the one after. They never got to vote, yet they must pay the bill for this transformation.
These three layers of fear overlay, combined with the fierce infighting within elite circles (acceleration vs. pause, Musk vs. Altman), appear to ordinary people as nothing more than “gods fighting while mortals watch.” When normal channels cannot express discontent, distorted forms of “voicing” emerge.
IV. The Solution Is Not a Blog Post, But Real Institutional Construction
In his blog, Altman reflected on his mistakes and called for the decentralization of power. These personal statements carry value.
But one Molotov cocktail has already proven: personal reflection and touching family photos cannot save this spreading crisis.
We need at least the following four serious institutional responses:
1. Transparency and Substantive Participation: Open algorithms, independent third-party audits, and mechanisms that give the public genuine binding power — not superficial consultation, but real sharing of authority.
2. Social Buffer Mechanisms: Large-scale retraining programs, transitional income support, and deep reform of the education system. Transform the majority from “those replaced by AI” into “effective users of AI.”
3. Balanced Governance Framework: Avoid excessive regulation that stifles innovation, while establishing mandatory external checks and balances. True “democratization of AI” must not remain forever a slogan.
4. Reducing Confrontational Atmosphere: Tech leaders, critics, and the media must all stop fueling zero-sum thinking. Anyone who continues to add fuel is helping light the next Molotov cocktail.
These suggestions are not new. But the Molotov cocktail has changed their urgency — from “should be done” to “must be done immediately.”
V. Conclusion: Do Not Wait for the Next Explosion
The biggest flaw in past AI discussions has been: elites talking to themselves while societal anxiety fermented in silence.
Now, the silence has been broken — in the ugliest and most dangerous way.
If we merely condemn the violence and then return to writing papers, posting tweets, and filing lawsuits, the next Molotov cocktail may not bounce off so fortunately.
Altman’s sharing of family photos is touching. But true safety has never been secured by a photograph. It comes from building an institutional track that allows everyone — including that 20-year-old attacker — to see hope for their own future.
Fear is spreading rapidly.
AI has arrived too fast, too powerful, and too unfairly. So fast that we must lay institutional tracks and build buffer zones. Otherwise, the next “explosion” will no longer be a metaphor.
This fire did not kill anyone, but it burned through a dangerous illusion and made one thing clear:
The future of AI can no longer be decided by a small minority alone.
