Cutting-edge artificial intelligence models are improving rapidly enough to outpace widespread cybersecurity know-how within months, the Five Eyes spy agency has warned.
In April, startup Anthropic announced that its cutting-edge Mythos model is Unprecedented ability to discover software vulnerabilities.
Security authorities in the UK, US, Australia, Canada and New Zealand called on governments and businesses to act quickly to prepare for advances in AI.
“The rapid pace of frontier AI development means that cyber risk assumptions could become outdated in months, not years,” they said in a joint statement on Monday.
The Five Eyes recommendation states that AI “lowers barriers to malicious actors and increases the speed and complexity of attacks.”
“Breaches will happen, and being prepared can help you contain them quickly and prevent them from escalating into a major operational or financial crisis.”
To strengthen cyber defenses, organizations need to integrate AI tools into their security operations, update outdated systems and restrict access to critical systems, he said.
“This is pretty alarming,” former U.S. Cybersecurity and Infrastructure Security Agency Administrator Chris Krebs told CBS News.
“The past few months have been a bit of a whirlwind in terms of advanced AI,” he said. “And this memo from the Five Eyes intelligence agency, while not really a single development, is a signal that businesses need to take very seriously the risks posed by AI falling into the wrong hands.”
Krebs said the Five Eyes statement “suggests a number of discrete and achievable actions” that businesses and other organizations can take to “make themselves a tougher target, more agile, more resilient” and prepare for what he called “the tsunami of vulnerabilities that are coming our way.”
Anthropic earlier this month suspended access to restricted versions called Mythos 5 and Fable 5 to comply with U.S. national security orders.
Just days after the launch of Fable 5, the company announced that it had received a government order barring all foreigners from accessing the two models.
The intervention is shocking for a White House that has pushed for less oversight of AI and even tried to block states from creating their own rules.
