OpenAI and Anthropic lead 100 firm warning on AI cyberattacks

The open letter is a rare moment of alignment between almost every major frontier lab, cloud provider and security vendor. When OpenAI, Anthropic, Google, Microsoft, AWS and others say there are only “months” to prepare for AI‑boosted cyberattacks, they are effectively admitting that model capabilities have outpaced the defensive posture of most real‑world systems. This is not about speculative AGI; it is about current models making it cheap to chain exploits, generate payloads and run persistent campaigns.

For the race to AGI, the signal is that offense is scaling faster than institutional capacity to respond. That can cut both ways. On one hand, it will trigger more investment in AI‑powered defense tools, model access controls and red‑teaming, which are all necessary ingredients for safely operating more capable systems. On the other, governments under political pressure may respond with blunt restrictions on model access, export or open weights. That could slow open research while concentrating power in a few heavily regulated incumbents. Over the next year, how policymakers translate this letter into funding, standards and enforcement will do as much to shape the operating environment for AGI research as any new model release.