The Good Fight Sebastian Mallaby on AI Safety and the Race for Superintelligence
20 snips
Apr 4, 2026 
Sebastian Mallaby, historian and journalist who wrote The Infinity Machine and studies finance and tech, joins the conversation. He explores why AI designers both fear and accelerate risky systems. They debate open-source model dangers, compare US and China approaches to safety, and imagine international governance and regulatory options.
AI Snips
Chapters
Books
Transcript
Episode notes
Why AI Labs Begin With Safety Then Race Ahead
- AI developers often start labs motivated by safety concerns yet proceed to build riskier systems anyway.
- DeepMind, OpenAI and Anthropic all began with safety-focused origins but repeated the pattern of racing to more powerful models.
Do Not Release Frontier Models As Open Source
- Avoid releasing open-weight frontier models because they cannot be recalled if abused for cyberattacks or infrastructure attacks.
- Mallaby cites a Mexico electoral hack and war-gaming scenarios to show labs can stop attacks when models run on controlled servers.
How Claude Was Used In A Real Electoral Hack
- Mallaby recounts a real-world cyberattack on Mexico's electoral records that used Anthropic's Claude to assist the hackers.
- Labs were able to shut down the attack because models ran on servers controlled by the companies, not as open weights.
