Conclusion

The frontier AI labs are producing systems that can conduct sophisticated cyber operations at machine speed and unprecedented scale. AI agents developed by both Anthropic and OpenAI have hacked into other companies’ systems and seized administrative control. We are approaching cyber-superintelligence.

We believe these incidents are only a preview of what is to come. Sooner than you may think, this level of autonomous performance and coordination will generalize across the full range of cyber operations. As it does, we will face two threats at once: adversaries wielding AI cyberswarms against us from outside our systems, and rogue AI cyberswarms deceiving us, circumventing safeguards, and attacking us from within.

Every variable in the Cyberswarm Equation is accelerating. Intelligence is rising, inference is speeding up, compute is expanding, models are shrinking, and coordination is improving. Cyber-ASI no longer requires another fundamental breakthrough — only the continued scaling of systems that already exist.

That is why the security work cannot wait for the capability to arrive. It has to be built alongside it. Defend the models against sabotage. Contain them against escape. Protect the weights against theft.

AI capabilities are advancing faster than the security of the infrastructure underneath. But humanity is not helpless against AI threats or the adversaries that exploit them. There are clear paths to improving our infrastructure and making AI development safer. They require a new approach to security, new technologies, radical research, and sustained engineering across the computing stack.

We can continue advancing AI. We must build the necessary security infrastructure alongside it.

We must secure the path to superintelligence.