Multiple things can be true at the same time

A security professional at Mozilla reflects on the tension between deeply criticizing AI (for its environmental costs, bias, plagiarism, erosion of trust in writing, and intensification of work) and pragmatically using LLMs in security work. The author describes how AI-assisted red-teaming with Anthropic uncovered 14 bugs in Firefox, which then led to finding and fixing 271 more, including sandbox escapes. The conclusion is that holding contradictory truths simultaneously is necessary: AI has serious harms, yet security practitioners cannot afford to ignore tools that provide a meaningful advantage over attackers.