Five Eyes agencies warn AI models pose growing cybersecurity risks, urge faster defense adaptation

National security agencies from the Five Eyes alliance—the U.S., U.K., Canada, Australia, and New Zealand—jointly released a statement last week warning that AI models pose increasing cyber risks, particularly their ability to autonomously hack into systems and networks. The agencies described the risks as accelerating due to the widening gap between skill and ability enabled by AI tools.

The statement clarifies that while AI expands the scope of what users can accomplish, it also lowers the technical barriers to sophisticated attacks. Historically, cyberattacks required significant expertise, but AI tools—including smaller, open-source models—are increasingly capable of executing attacks autonomously with minimal prompting. This shift mirrors past trends in hacking tools, where prewritten exploits democratized attack capabilities for less skilled actors.

The Five Eyes agencies argue that the rapid pace of AI development means cyber risk assumptions can become outdated within months, not years. They recommend adopting AI-driven defenses to detect vulnerabilities earlier, improve software quality, monitor unusual behavior, and respond faster to incidents. The agencies acknowledge that their advice is not new but emphasize its renewed urgency in the AI era.

The essay accompanying the statement highlights the dual-use nature of AI capabilities. While AI can strengthen cybersecurity by automating defensive tasks, the same knowledge used to build defensive tools can also be repurposed for offensive purposes. For example, AI models capable of reviewing code for vulnerabilities can also be instructed to exploit those vulnerabilities. This creates a persistent challenge: restricting harmful uses without crippling beneficial applications.

The agencies also note that guardrails implemented by major AI developers to prevent misuse are unlikely to be effective long-term, especially as smaller, open-source models—often run locally—circulate without such constraints. They suggest that the most viable path forward is to harness AI for defense while preparing for a world where both beneficial and harmful actions are supercharged by AI assistance.