The AI Safety community is unfortunately doing more harm than good
Short summary
A tech worker argues that recent public statements from AI lab employees about existential risks—including Palisade Research interviews and Jacob Coxon's departure from Anthropic—are harming the AI industry by fueling anti-AI sentiment among ordinary Americans rather than advancing serious safety discourse. He contends these warnings play into the hands of politicians and activists who want to ban AI entirely, without reaching the people who actually need to understand alignment research.
- Jacob Coxon's post about Anthropic moving too fast toward superintelligence reached 150+ million views and went mainstream on TikTok and YouTube, polarizing AI development into a partisan issue between accelerationists and pausers.
- Average Americans opposing AI aren't making informed decisions—they're reacting to data center construction near their communities and messaging from politicians, not engaging with technical safety literature like alignment research or essays on AI governance.
- Public doomer messaging from current AI lab employees (claiming "human extinction is a coin flip") alienates the general public and strengthens the case for AI regulation or bans, rather than building support for the industry's continued development.
Full summary
The author argues that recent warnings from AI lab employees—particularly Jacob Coxon's departure from Anthropic and Palisade Research's interviews with 22 current and former AI workers—are backfiring politically. Coxon's post about labs "gambling with our lives" hit 150 million views and went mainstream through TikTok and YouTube. Before that, 1,386 employees signed the July 2026 "Pacing the Frontier" letter demanding AI labs slow down. While these warnings come from genuine safety concerns, the author sees them as handing ammunition to actual AI opponents in Congress and the general public who want outright bans, not thoughtful regulation.
The real damage happens at the perception level. When average Americans—people not reading LessWrong or following model releases—see researchers making millions saying extinction is a coin flip, they don't engage with nuanced safety discussions. They get angry at tech executives and want AI banned. The author notes most people aren't informed about things like METR (which evaluates models before release) or why data centers exist. They just hear "dangerous AI" from authority figures and assume it's true. Nobody outside the tech bubble reads the 5,000-word rebuttals from people like Andy Masley that dismantle these claims one by one.
The author's core complaint: this messaging is tactically stupid. It won't convince AI accelerationists to pump the brakes, and it actively mobilizes the anti-AI crowd to share doomism in group chats and demand bans. He understands the safety concerns are real and that guardrails matter, but believes the current approach of platforming high-doom scenarios will trigger actual legislation banning AI development entirely. The internal fighting over whether to accelerate or pause is obscuring what might actually matter—getting ahead of a political backlash that could kill the entire industry.