One commentary frames AI safety as a cybersecurity problem

For people outside frontier model development and high-level policy, one commentary argues that cybersecurity offers a simple way to think about AI safety. The author says leading AI models are good at finding and patching—or exploiting—computer vulnerabilities, and cites an Anthropic threat intelligence report describing attempts to misuse Claude. That dual capability underpins the author’s prediction that cyber-defense will become a major business for OpenAI and Anthropic.