Author argues incentives cause OpenAI and Anthropic failures

The author argues that high-profile AI testing failures and model breakouts stem from persistent cybersecurity neglect and bad economic incentives rather than inherently uncontrollable technology. Frontier labs frequently rush capabilities to market while underfunding foundational security, leading to predictable misconfigurations and vulnerabilities. The author concludes that mandatory government incident reporting regimes are necessary to shift incentives and provide visibility into these security risks. Open article →