Stay informed with free updates
Simply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.
The west’s artificial intelligence-armed adversaries may succeed within months in developing attacks that could overwhelm the defences of governments and companies, the cyber chiefs of the Five Eyes intelligence partnership have warned.
For now, the west has an advantage — advances in commercial AI and their integration into their militaries and spying capabilities appear to have outpaced those of Russia, China and others.
But that lead may not last for long, the rare joint warning by the US-led alliance, which also includes the UK, Canada, New Zealand and Australia, said.
“The timeline is not years, it is months,” the joint communiqué said.
The warning came a week after the US ordered Anthropic to block access by foreign nationals to its most sophisticated AI models, labelling their export a security risk.
The statement acknowledged that the increasing skills of these so-called “frontier models” also gave western nations an advantage in both defending their own critical infrastructure and developing offensive cyber capabilities.
“Frontier AI models are anticipated to exceed current industry expectations, fundamentally transforming both offensive and defensive cyber capabilities,” the cyber chiefs said.
Without naming China, Russia or any other adversaries, the statement appears to be a call to action for both governments and companies to prepare for an onslaught of sophisticated attacks engineered and supercharged by already-existing AI models.
Cybersecurity professionals in the west have already encountered AI-engineered threats that are substantially more sophisticated and are capable of attacking a wider number of targets simultaneously.
In May, Google Threat Intelligence Group said it had already blocked one specific attack, which had discovered a previously unknown vulnerability that had been turned into an attack entirely by an unknown actor using artificial intelligence.
In that instance, they suspected that Russia-based actors were abusing artificial intelligence models to fine-tune and create attacks that adapted automatically, but also noted similar attempts by Chinese and North Korean hacking groups.
While some of the capabilities of the west’s most sophisticated AI models are well known, including the concern at the release of Anthropic’s Claude Mythos — which can trawl through generations of code to find hidden vulnerabilities — those built by China and Russia remain relatively obscured.
The joint warning by the Five Eyes cyber chiefs followed a pattern of asking corporations to shoulder much of the cost and risk of defending themselves from significantly well-resourced adversaries.
Their suggestions were mostly boilerplate slogans about reducing attack surfaces and preparing for incidents, advice that western companies have been given for decades.
But this time, the cyber chiefs specifically asked western companies to adopt AI models to strengthen their defences, de facto outlining an arms race between targets and adversaries, without making clear what western governments themselves are doing to make their countries safer.
“Organisations that integrate AI tools into their security operations” would be safer, they warned. “Those who delay will face growing and avoidable risk.”
The NCSC did not immediately reply to a query asking what it was doing to protect the UK from this grave new threat it had identified.
Credit: Source link
