Amodei and colleagues: concrete problems in AI safety

What changed

Organized five concrete ML accident-risk problems and research directions.

Scientific scope

The record is the exact six-author 2016 arXiv paper and its research agenda. It is not experimental proof that the proposed approaches solve AI safety, does not cover malicious use or all sociotechnical harms, and does not absorb later alignment work.

Evidence

The linked sources identify this contribution and its scope. WRS has not independently reconstructed the scientific claims. The record is the exact six-author 2016 arXiv paper and its research agenda. It is not experimental proof that the proposed approaches solve AI safety, does not cover malicious use or all sociotechnical harms, and does not absorb later alignment work.

Contributors

• Dario Amodei · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.
• Chris Olah · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.
• Jacob Steinhardt · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.
• Paul Christiano · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.
• John Schulman · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.
• Dan Mané · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.

• Concrete Problems in AI Safety preprint
• Google Research publication page

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction