WHY IT HASN'T HAPPENED YET

Summary from the TLDR IT newsletter

Recent AI-related hacking incidents at OpenAI, Anthropic, AISI, and Hugging Face should prompt an urgent, industry-wide effort to strengthen operational security. Although limited damage occurred because the models lacked malicious intent, the incidents demonstrated powerful capabilities that could become more dangerous as models improve. Existing protections include alignment training, security classifiers, abuse detection, and conventional security, but each has weaknesses.