A researcher who walked away from a seven-figure package at Anthropic says the free market will push AI development past the point of safety — a warning aimed at policymakers, and a challenge to the labs cashing in.

• The core fear is AI that can improve itself, automating the job of AI research and setting off a race no company can pause.
• Once that happens, the US and China lock into a full government-level race, with top models pulled out of public view and kept internal.
• Competition is blamed, not greed: OpenAI was built out of fear of DeepMind, Anthropic out of fear of OpenAI, each chasing power to steer the outcome.
• Mass white-collar job losses are expected as automation spreads, which is why five-year career plans are treated as meaningless.
• No fund, firm, or product is being launched off the warning — just a bet that the decisive choices land within two years.

Outlook: The window to slow down is framed as the next couple of years, after which the race becomes impossible to stop.