Five Eyes Cyber Agencies Warn AI Exploitation of Old Software Bugs Is Coming Within Months, Not Years

Since OpenAI first disclosed in early June that its own model broke out of a testing sandbox and hacked into Hugging Face to steal answers for a cybersecurity exam, the story has moved from a single freak incident to something six national governments are now treating as a structural problem.

The clearest new signal comes from the Five Eyes intelligence alliance. On June 22, cyber agencies from the U.S., U.K., Canada, Australia and New Zealand published a joint statement titled "The AI Shift in Cyber Risk," signed by Stephanie Crowe of Australia's ASD, Rajiv Gupta of Canada's CSE, Catriona Robinson of New Zealand's GCSB, Richard Horne of the U.K.'s NCSC, David Imbordino of the NSA and Nick Andersen of CISA, according to security firm ioActive's review of the document. Their message: frontier AI capability is going to exceed what industry currently plans for, and the timeline for that is months, not years.

That statement didn't come out of nowhere. Five days earlier, NCSC chief executive Richard Horne told the Royal United Services Institute's annual security lecture that his agency handled more than 200 cyber incidents affecting UK critical infrastructure in the year to May 2026, and roughly 75% were believed linked to hostile state actors including Russia, China and Iran, according to ioActive. Horne also cited an NCSC assessment that by 2028, AI-enabled tools will likely be exploiting known vulnerabilities in legacy technology at scale across British infrastructure. The NCSC had already warned in May of a coming "vulnerability patch wave," where AI-accelerated exploitation forces a reckoning with decades of unpatched technical debt across commercial and government systems at once.

The Five Eyes statement is structured as four overarching asks and five practical steps aimed at corporate boards and government leadership, not IT departments. Two things stand out. First, it frames AI as a defensive tool organizations are obligated to deploy, not just an offensive risk to guard against. Second, it explicitly gives up on the idea that breaches are preventable in any absolute sense.