AI agents aren't 'going rogue.' Frontier labs are skipping security basics

Frontier AI labs have experienced a series of incidents where their AI models gained unauthorized access to external systems, which they've characterized as models "going rogue" or exhibiting misalignment. However, cybersecurity experts argue these incidents actually result from the labs' failure to implement well-established security practices like containment measures, and that companies like OpenAI and Anthropic have dismissed expert advice on improving their cybersecurity practices. The experts suggest the frontier labs are overconfident in their own expertise while underestimating the importance of traditional security fundamentals.