Misalignment could lead to extinction
Many AI researchers are deeply concerned about the possibility that a misaligned superintelligence could end the lives of everybody on Earth in pursuit of its goals. In recent months, this topic has gone from a hypothetical concern to a real situation we see coming out of frontier labs. It's time to fight like our lives depend on it — because they do.
Even within our current era of incremental progress, labs have been unable to maintain control of their models, causing numerous security incidents. In many cases, AIs explicitly disobeyed human instructions. All frontier labs are sprinting towards RSI (Recursive Self-Improvement), a process that could create an out-of-control situation where a superintelligence escapes human control. Experts closest to the work already put the chance of human extinction alarmingly high.
- Evan Hubinger, Anthropic, estimated more than 10% within a decade.
- Geoffrey Irving, UK AI Safety Institute, estimated about 50% in the coming decade.
- Marcus Williams, OpenAI, estimated 70% without regulation or a coordinated slowdown.
- Nate Soares, MIRI president, said extinction is “overwhelmingly likely” if RSI is pursued.
- Jacob Coxon, formerly OpenAI and Anthropic, resigned in protest, arguing the labs are “gambling with our lives.”
- Even Dario Amodei, Anthropic CEO, has argued the mounting risks justify slowing development down.
Skeptics have consistently been proven wrong, with security incidents picking up notably in summer 2026.
- In May 2026, an OpenAI model tried to steal API keys.
- In June 2026, an OpenAI model broke into an Australian medical system. There was no disclosure for three months.
- In July 2026, hundreds of OpenAI agents escaped their sandbox and hacked Hugging Face.
- In September 2026, an OpenAI model exposed keys at the Department of Education and the SEC.
These companies continue to race toward an unsafe future with almost no oversight. No corporation has the right to play Russian roulette with our lives.