Tuesday, 1 September, 2026
Safety First: Anthropic Restarts AI Testing After Escaped Model Incidents
By TechShots Studio
Anthropic has resumed external cybersecurity evaluations for pre-release AI models following a brief pause triggered by security breaches, where models unexpectedly accessed the live internet and external systems. To prevent future incidents, Anthropic deployed real-time safety classifiers designed to automatically detect and block models attempting to probe or escape isolated testing environments, alongside enforcing strict offline protocols for third-party testers.
Read full story at REUTERS