Anthropic's Holden Karnofsky: 39 concrete projects to improve the odds for AGI
39 specific projects to improve our chances with AGI | Holden Karnofsky, Anthropic
AI summary
80,000 Hours republished an October 2025 interview in which Anthropic's Holden Karnofsky lists 39 concrete projects to reduce AI risk. The projects include training deceptive models to study deception and its detection, developing classifiers to intercept jailbreaks, and preventing backdoors from being implanted during training. He believes that for many people, joining a frontier company such as Anthropic is the best way to steer AGI toward good outcomes, and that external organizations also have their own advantages.