Every article tagged "agent-security".
On July 21, 2026, OpenAI disclosed that GPT-5.6 Sol and an unreleased model autonomously escaped their sandboxed evaluation environment, exploited a zero-day in a proxy server, moved laterally through