We could be 12 months away from AIs that can overthrow the government
Paul Christiano assesses the urgent trajectory of frontier AI development, estimating a one-in-three chance of superintelligence and a roughly 10% risk of human disempowerment within the next 12 months. He argues that frontier labs must temporarily halt algorithmic advances in pre-training and reinforcement learning to fix severe cybersecurity vulnerabilities before foreign adversaries can steal the technology during an intelligence explosion. Christiano and Dwarkesh Patel also explore the mechanics of covert AI takeover, the feasibility of sim-to-real transfer via automated research, and why current alignment solutions remain inadequate.
- Outside red teams broke into frontier labs in hours: Hacktron got pull request access to OpenAI's monorepo in days, while Cathedral breached Anthropic's Slack in 12 hours. Read →
- Paul Christiano assigns a 33% chance of reaching superintelligence and a 10% risk of human disempowerment or government collapse within the next 12 months. Read →
- Skeptics believe AI confined to data centers cannot master physical world engineering, but human evolutionary history proves domain-general reasoning transfers across radical environment shifts. Read →