AI models capable of launching large-scale cyberattacks that could overwhelm government and corporate defenses are months, not years, away, the international coalition of intelligence agencies warned in a joint statement.
The Five Eyes, made up of the US, UK, Canada, Australia and New Zealand, called on government and business leaders to “act now” to strengthen defenses against advanced cyber threats.
The unusual call to action comes after the Trump administration ordered artificial intelligence giant Anthropic to stop foreigners from using its cutting-edge models, underscoring growing anxiety in the West about the technology’s new capabilities.
“The Frontier Al model is expected to exceed current industry expectations and fundamentally transform both cyber attack and defense capabilities. The timeline is months, not years,” the spy agency group said in a statement Monday.
“Cyber risks are changing rapidly due to advances in artificial intelligence (AI), and we need to act quickly to stay ahead.”
AI researchers and executives have raised a variety of security concerns about this advancing technology, with Five Eyes leaders explaining that the technology could “lower barriers for malicious actors and increase the speed and complexity of attacks.”
AI experts said the message was “truly harsh” and could have alarming implications for governments and businesses, as well as small and medium-sized businesses around the world.
“In the age of AI, breaches are going to happen. It’s not a question of if, it’s not a question of when. So it’s important to prepare now,” Olivia Shen, director of the Strategic Technology Program at the Center for American Studies at the University of Sydney, told CNN.
The U.S. government’s wide-ranging directive for Anthropic’s Mythos 5 and Fable 5 models was one of the most far-reaching actions the government has taken in response to the advanced capabilities of AI models.
Mythos had sparked widespread cybersecurity concerns as it said it was very adept at finding security flaws. But Anthropic said it believes the U.S. government has “realized” a way to “jailbreak” the public Fable model, or circumvent domestic security guardrails. Anthropic and the administration are meeting to resolve the issue.
“The really important takeaway from this is that AI capabilities are evolving incredibly rapidly,” Shen said, adding that while the world’s attention is currently on Anthropic, someone else could create the next highly capable AI.
To combat this threat, Five Eyes leaders said companies and leaders must invest in cyber defenses, upgrade outdated systems or patch flawed software and restrict who has access to critical systems.
They added that AI, which is being used by adversaries to “act faster and more effectively,” is also part of the solution.
“Organizations that integrate AI tools into their security operations will be able to detect vulnerabilities early, improve software quality, monitor anomalous behavior, and respond quickly to incidents,” the Security Alliance said.
Shen, an expert on AI and national security, said there are “huge gaps” in the defense capabilities of many governments and companies.
“Sophisticated companies, usually larger companies, have already invested in cybersecurity and will be better prepared,” she said. “What’s more at risk are probably the smaller businesses that have been under-invested so far, and they’re basically going to be sitting ducks.”
AI models are advancing at breakneck speed, with independent assessments showing that some models have reached expert levels of cyber capability. The pace appears to be giving lawmakers a losing battle as they try to install guardrails.
Currently, the United States lacks a transparent and consistent framework for regulating AI. Some experts say the government should be involved in the debate about AI safety, while others say the outcome could stifle the industry.
Dozens of cybersecurity researchers, AI entrepreneurs, and business executives signed an open letter this month calling on the Trump administration to engage in “an open, scientific, and transparent process to address AI risk assessment,” saying it is “essential” that security teams “find and fix flaws in their newly written code and decades of legacy code before their adversaries do.”
Mr. Shen acknowledged there was a “tightrope to walk” but said some “ground rules” were needed.
“We know these technologies can be used for both defensive and offensive purposes, and we need some more guardrails on how to maximize the benefits of defensive cybersecurity while keeping cybersecurity away from potential cyber adversaries, fraudsters, and cybercriminals,” she said.
