AI cybersecurity risk could outpace current defences within months, the Five Eyes intelligence alliance warned in a joint advisory.
The warning came from security agencies in the United States, the United Kingdom, Australia, Canada and New Zealand. They urged governments and businesses to prepare for faster and more complex cyberattacks.
The advisory said frontier AI development could make current cyber risk assumptions outdated “in months, not years.”
The Five Eyes advisory said artificial intelligence can lower barriers for hackers. It also said AI can increase the speed and complexity of attacks.
The agencies warned that breaches would happen. However, they said strong preparation could help firms contain attacks before they grow into major crises.
The agencies urged organisations to incorporate AI tools into their security operations. They also advised firms to update old systems and patch software flaws.
Access to critical systems should be limited, the advisory said. The warning followed concern over Anthropic’s advanced Mythos models.
Read: Pakistan Mandates Cybersecurity Training in Schools to Strengthen Digital Safety
In April, Anthropic said its models demonstrated a strong ability to detect software flaws. This month, Anthropic suspended access to Mythos 5 and a restricted model called Fable 5.
The company said it acted in response to a US national security order. The directive barred foreign nationals from accessing the two models.
Meanwhile, the US cyber-defence agency CISA shortened federal vulnerability response deadlines to three days. The agency cited AI-related threats.
