Anthropic has published research on an AI-driven system called Co-RedTeam, built to discover and exploit software security ...