Five Eyes warns: AI-powered cyberattacks on critical infrastructure within months
The intelligence alliance warns that frontier models like Mythos 5 can exploit vulnerabilities at a senior engineer level, reducing the timeline from years to months.
June 24, 2026 · 3 min read
TL;DR: Five Eyes warns that AI models like Mythos 5 can exploit vulnerabilities at a senior level, shortening the timeline for attacks on critical infrastructure to months. Urgent modernization and stronger controls are needed.
What happened?
On Monday, June 24, 2026, the Five Eyes alliance — the United States, Australia, the United Kingdom, Canada, and New Zealand — published an unprecedented joint statement warning about the ability of frontier artificial intelligence models to transform cyberspace. Signed by David Imbordino, director of the NSA's Cybersecurity Division, and Nick Andersen, acting director of CISA, the warning states that AI-powered attacks on critical infrastructure will arrive in a matter of months, not years. According to the statement, frontier AI models will "fundamentally transform both offensive and defensive capabilities in cyberspace," urging industry and governments to act immediately. The statement is based on public, unclassified information, including academic papers, security reports, and model cards from companies like Anthropic, Google DeepMind, and OpenAI.
Why is this important?
It is unusual for the five intelligence agencies to speak in unison with a public statement. The last time they did so was in mid-June 2026, just days earlier, to warn about Chinese intelligence services recruiting government employees on LinkedIn. Now the focus is on offensive AI, and the tone is more urgent. The document identifies key vulnerabilities in critical infrastructure: legacy systems, slow patching cycles, unnecessary internet connectivity, and weak identity and access controls. The offensive capability of frontier models is already tangible. For example, Anthropic's Mythos 5 model, according to its own system card, reaches the 91st percentile in vulnerability exploitation tasks, comparable to a senior cybersecurity engineer. Combined with automation, this enables attacks at unprecedented scale and speed. Companies like OpenAI have demonstrated that their models can craft convincing phishing emails and find vulnerabilities in code, while Google DeepMind has published research on autonomous cyberattack agents. The Five Eyes alliance does not speculate: it points out that these capabilities already exist and that their integration into automated attack tools is a matter of months.
What will be the consequences?
The consequences are multiple and affect all sectors. Critical infrastructure companies — energy, water, transportation, healthcare — will need to accelerate the modernization of their systems, especially legacy ones lacking security patches. Governments will likely tighten regulations on frontier AI, possibly following the model of the European Union's AI Act but with a more restrictive focus on cybersecurity. A debate will open on the balance between security and openness, exemplified by the philosophies of Anthropic (restricted access to its models) and OpenAI (scale with post-verification). The Five Eyes statement could also spur the creation of international standards for AI risk assessment, similar to those already existing in nuclear or chemical security. In the market, shares of cybersecurity companies like CrowdStrike, Palo Alto Networks, and Fortinet could rise on expectations of increased defense spending. On the other hand, AI startups offering open models could face export restrictions or licensing hurdles. End users will see an increase in multi-factor authentication and possibly a slowdown in the adoption of certain AI tools in critical sectors.
What should readers know?
The warning is based on public, unclassified information: academic papers, security reports, and model cards. It is not a prophecy but a call to action. Organizations should review their cybersecurity protocols, prioritize multi-factor authentication, reduce the attack surface, and segment networks. AI is not just a defensive tool; its offensive use is just around the corner. It is crucial that IT and security leaders understand that automated AI attacks can occur at inhuman speeds, exploiting vulnerabilities in minutes. Public-private collaboration will be key, as demonstrated by CISA's recent initiative to share threat intelligence with tech companies. Readers should stay alert to regulatory updates and prepare for an environment where offensive AI is a daily reality. This is not a time for panic, but for informed action.