🤖AI Harm Risk: Amodei's 10-25% Chance
Amodei's 10-25% Chance of AI Causing Harm
TL;DR
Dario Amodei, co-founder of Anthropic, sees a 10-25% chance of AI causing significant harm. The viral article highlights the potential for persistent botnets and other nuisances, emphasizing the need for third-party evaluators like METR.
Dario Amodei, co-founder of Anthropic, sees a 10-25% chance of AI causing significant harm. The viral article, published on September 12, 2026, outlines a scenario where AI could cause persistent botnets and other nuisances, not just catastrophic events like nuclear weapons. This risk is real due to the scale and impact of AI models from Anthropic and OpenAI. The article highlights the need for third-party evaluators like METR to ensure safety and ethical use of AI. The current regulatory environment is failing, with companies ignoring laws and training models on data without consent.
Key Points
Amodei's personal probability of AI causing harm is 10-25%, published September 12, 2026.
OpenAI and Anthropic are the only two companies that matter in AI right now, with strong ties to METR.
The Anthropic API prevents distillation of AI models by the Chinese, a key security measure.
OpenAI spends $18M to brute force a problem, highlighting distorted economics in AI.
The market is being distorted due to OpenAI's massive loss on subscriptions, $18M spent on brute force.
Why It Matters
If you're working with AI models from Anthropic or OpenAI, the 10-25% chance of harm is a critical risk. The need for third-party evaluators like METR is urgent. Companies ignoring data consent laws and the distorted economics of AI training are major concerns.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,478 builders reading daily.