Advanced artificial intelligence models are evolving rapidly enough that assumptions underpinning cybersecurity defenses may become outdated within months rather than years, according to a warning issued Monday by an alliance of intelligence and cybersecurity agencies from the United States, Britain, Australia, Canada and New Zealand.
The joint advisory from the Five Eyes alliance urged governments and businesses to prepare for a changing threat landscape by modernizing systems, strengthening access controls and incorporating AI tools into their security operations.
The agencies said cyber resilience should be treated as a leadership priority and noted that organizations should expect breaches to occur and focus on limiting damage and restoring operations quickly.
The warning comes at a time when cyberattacks have become increasingly intertwined with geopolitical conflicts. Governments and security firms have documented hacking campaigns tied to the war in Ukraine, while Western countries have also accused Iranian-linked groups of conducting cyber operations during heightened tensions in the Middle East. Critical infrastructure, government networks and private companies have remained frequent targets.
The Five Eyes said advances in frontier AI systems are reducing barriers for malicious actors while increasing the speed and sophistication of attacks. Existing assumptions about cyber risks can become obsolete much faster because of the pace of AI development, according to the statement cited by Reuters.
The concerns have gained urgency following recent developments involving AI giant Anthropic. In April, the company announced a research initiative involving its Mythos family of models, which demonstrated unusually strong capabilities in identifying software vulnerabilities. The systems uncovered thousands of security flaws across operating systems and browsers, according to TechCrunch.
Anthropic later said it had disclosed more than 1,500 vulnerabilities affecting hundreds of open-source software projects. The company presented the effort as part of a coordinated vulnerability disclosure process intended to improve software security.
Attention around the models intensified this month after Anthropic suspended access to Mythos 5 and a restricted version called Fable 5 in response to a U.S. national security directive. The company said it had received an order prohibiting foreign nationals from accessing the models and complied with the request while maintaining that officials may have misunderstood aspects of the technology, according to a statement published by Anthropic.
Concerns about the pace of AI development have also emerged within intelligence circles. Western security agencies fear that longstanding assumptions about cyber defense no longer reflect the speed at which advanced AI systems are evolving, Financial Times reported.
The White House has generally favored a lighter regulatory approach toward artificial intelligence and has sought to accelerate U.S. development in the sector. At the same time, debates over oversight have intensified as policymakers weigh national security concerns against efforts to maintain American competitiveness in AI.
The warning highlights growing attention on cyber threats at a time when ransomware attacks, state-backed hacking operations and digital espionage campaigns continue to affect governments and businesses worldwide. Recent incidents involving healthcare providers, telecommunications firms and critical infrastructure operators have underscored concerns about vulnerabilities in interconnected systems.
