WASHINGTON — The intelligence agencies of the United States, Britain, Canada, Australia and New Zealand issued a rare joint statement Tuesday warning that a new generation of artificial-intelligence models could break through government and corporate cyber defenses “in a matter of months, not years,” according to people who reviewed the statement. The Five Eyes alliance, which coordinates intelligence sharing among the five English-speaking democracies, said the threat has moved from a distant concern to a near-term planning problem.
The statement, published simultaneously by the five agencies, said AI models are now capable of automating attacks that previously required skilled human operators, including the discovery of software vulnerabilities, the crafting of convincing phishing messages and the evasion of detection systems. Reuters and CNN both carried the warning as a lead story, a measure of the gravity the agencies were seeking to convey.
The compressed timeline is the striking part. Previous assessments from Western governments had described AI-enabled offensive capabilities as emerging within years, with human oversight required for the most dangerous operations. The new statement collapses that estimate to months, citing rapid advances in model capabilities and the availability of open-weight models that any actor can adapt for malicious purposes.
The warning reflects a shift in how intelligence agencies think about AI. For the past two years, the dominant concern was defensive: protecting AI systems themselves from attack and theft. The Five Eyes statement broadens the frame to include AI as an offensive tool, one that lowers the skill barrier for cybercrime and espionage alike. A single operator with a capable model and a rented botnet can now attempt attacks that once required a team, the agencies said.
The implications for organizations are immediate, security analysts said. Companies that have treated cyber defenses as a compliance exercise will need to assume that attackers have AI assistance around the clock, and that the volume of attacks will rise even if the sophistication of individual attacks does not. Patching cycles, which already struggle to keep pace with known vulnerabilities, will come under additional pressure as AI models discover new flaws faster than vendors can fix them.
The statement also carries implications for the private AI companies whose models make the warning possible. The agencies did not name specific companies, but security researchers have documented that frontier models can write working exploit code and that open-weight models have been used in real attacks. The Five Eyes governments have urged AI developers to build security testing into their release processes, and the statement said the agencies stand ready to work with companies that want to harden their models against misuse.
The warning is being met with some skepticism in the security community. Researchers who have studied AI-assisted attacks say the technology amplifies capability rather than replacing it, and that the most effective attacks still depend on human judgment and access. But the skeptics agree on the direction of travel: AI is making attacks cheaper, faster and more numerous, and the gap between model capability and defensive practice is widening.
For governments, the response is organizational as much as technological. The Five Eyes statement was coordinated through the alliance’s cyber operations centers, and each member government has announced plans to expand its cyber workforce and to fund AI-specific defense research. The warning’s timing, months before budget cycles close in several member countries, suggests the agencies are seeking additional resources with the urgency that a formal threat assessment provides.
The private sector is being asked to do its part. The statement encourages critical infrastructure operators to share threat information with government agencies, and several governments have proposed regulations requiring incident reporting within tight deadlines. Industry groups have pushed back on the reporting requirements, but the Five Eyes warning gives regulators a stronger hand in the argument over how much visibility companies must provide.
The response is already taking shape in national capitals. The five governments have announced expansions of their cyber forces, with Britain pledging new funding for its National Cyber Security Centre and the United States directing its agencies to accelerate AI-specific defense programs, according to officials familiar with the plans. The statements were coordinated, and the agencies have agreed to share threat intelligence on AI-assisted attacks through their existing channels, an arrangement that has been tested in past incidents. The coordination is itself a message: the intelligence alliance that was built for the Cold War and reshaped for the war on terror is now being pointed at a threat that does not respect borders, and the speed of its response will be measured in months, not years.
The warning has also reopened the debate over open-weight models. The agencies did not call for restrictions, and their statement emphasized that AI’s offensive potential is being realized by many actors, not just states, which points to a problem that export controls cannot solve. Security researchers have argued for more careful vetting of model releases, while open-source advocates say the benefits of open models outweigh the risks and that restricting them would cede the field to adversaries who will build them anyway. The Five Eyes statement does not settle that debate, but it gives regulators who want restrictions the strongest official assessment yet to cite, and the coming months are likely to see the argument play out in hearings and rulemakings on both sides of the Atlantic.
The central question, as one former intelligence official put it, is whether defensive technology can keep pace with offensive capability. The Five Eyes statement does not answer it; it merely insists that the race is being run now, and that the finish line has moved closer. Organizations that treat the warning as alarmism, the official said, will learn otherwise the first time an AI-assisted attack finds a hole in their defenses.
