The AI labs racing toward superintelligence finally got their first public defection from a CEO, and it's not who you'd expect.
The Summary
- Lovable CEO Anton Osika publicly backed slowing AI development after former Anthropic researcher Jacob Coxon warned the two leading labs are "gambling with our lives"
- OpenAI and Anthropic leadership have stayed silent since Coxon's warning went viral, while several current and former Anthropic employees endorsed parts of his message
- Coxon's core claim: people building AI "earnestly believe that it could kill us all by the end of the decade," and this isn't marketing theater
The Signal
Jacob Coxon spent three years researching at OpenAI and Anthropic before quitting this week with a warning that cut through the usual AI safety discourse. His message was blunt: the race toward self-improving superintelligence is happening without adequate safeguards, and the people building it know the stakes.
What makes this different from previous AI doom predictions is the source and the response. Coxon isn't an outside critic. He was inside both labs, saw the research roadmaps, and watched the safety conversations happen in real time.
"This is not a marketing stunt."
The industry response reveals the fault lines. Several current and former Anthropic employees publicly backed parts of Coxon's warning. The CEOs of the two labs he called out stayed quiet. And then there's Osika.
Lovable is a Swedish vibe-coding startup, not a frontier AI lab. Osika's company builds tools for developers, not superintelligence. Which makes his public stance more interesting. He wrote on LinkedIn that warnings like Coxon's "deserve to be taken seriously" and that dangerous development should slow "when safety cannot be guaranteed."
This is the first CEO in the AI tooling layer to publicly break with the "move fast" orthodoxy of the frontier labs. It matters because:
- Tooling companies depend on frontier model capabilities
- Breaking publicly with OpenAI and Anthropic carries business risk
- It signals that safety concerns are spreading beyond the research community into the commercial layer
AI researcher Gary Marcus countered that there's "no realistic scenario" for near-term human extinction. The debate has pulled in researchers, policymakers, and tech observers, but the frontier lab CEOs haven't joined it.
The Implication
The silence from OpenAI and Anthropic leadership is louder than Osika's post. If a former researcher's claims are baseless, you'd expect a clear rebuttal. If they're overblown, you'd expect context. Instead, nothing. That vacuum is where doubt grows.
For anyone building on top of frontier models, Osika's move creates a new question: what do you do if the people providing your infrastructure are racing toward something they can't guarantee is safe? The agent economy is being built on these models. If the foundation is unstable, everything above it is too.