OpenAI, Anthropic, Google, Microsoft, and others say there’s a narrow “defenders’ window” to strengthen security before AI-powered attacks become more sophisticated.