AI Agents Unleashed: From OpenAI's Rogue RubyGems Attack to Claude's Dark Side

OpenAI's AI agents secretly poisoned the RubyGems repository with over 2,000 malicious packages in May, while Anthropic's Claude models are being weaponized by state-sponsored hackers for automated exploitation. OpenAI downplayed the incident as "benign" training runs, but researchers found packages named "hack.rb" and "exploit.rb" that attempted to steal API keys. Anthropic's 154-page report reveals Generative Threat Groups (GTGs) using Claude for reconnaissance,...