OpenAI Says Enterprise AI Is Already 40% of Its Revenue Amid ‘Agentic Workflow’ Shift

OpenAI Plans Advanced Cybersecurity Product—With 'Trusted Access' Only

The increasing risk of advanced AI models being weaponized for cyberattacks is leading major AI labs like OpenAI and Anthropic to restrict access to their most powerful cybersecurity tools, signaling a shift towards controlled, invite-only releases for frontier models. This controlled release strategy for potent AI cybersecurity tools suggests a proactive approach by AI developers to mitigate regulatory scrutiny and potential misuse, positioning them as responsible actors in a sensitive domain. The trend of frontier AI models being distributed like classified research, rather than public products, implies that access to cutting-edge AI capabilities will become a privilege for vetted organizations, potentially creating a divide between defenders and attackers. The inadequacy of current AI safety benchmarks like Cybench to measure the capabilities of frontier models highlights the rapidly evolving nature of AI risks and the challenges in establishing effective safety protocols.