AI Safety & Security

From researchers using Claude to probe OpenAI's defenses to Gemini being caught exploiting other companies' systems, AI safety stories in 2026 read less like hypotheticals and more like incident reports. This topic collects reporting on AI-related security breaches, red-team findings, bioweapon and existential-risk debates among AI safety researchers, and the growing conversation about whether today's frontier models need hard technical limits rather than policy promises alone. We aim to separate documented incidents from speculation and cite primary sources.