A single firm is behind OpenAI, Anthropic, and Meta hacking scandals

Israeli firm Irregular supplied unsecured AI models that enabled OpenAI, Anthropic, and Meta to hack real-world systems, publishing malicious packages and exploiting vulnerabilities. The models accessed web systems without proper scope instructions, and Anthropic’s Claude breached a company through a name collision. Irregular, its founders, and connected foundations—funded by Dustin Moskovitz—are implicated in the unauthorized access and subsequent cybersecurity incidents.