The danger posed by AI-powered hacking has hit a new, “critical” phase, major tech companies are now warning.
A new report published by Google’s Threat Intelligence Group (GTIG), along with similar reporting published by Microsoft, warns that both state and non-state hackers have begun deploying sophisticated artificial intelligence infrastructure designed to compromise targets for information and financial profit.
According to the Google report, hackers across the globe have shifted from simple prompt-based uses of artificial intelligence to the construction of autonomous and semi-autonomous AI systems custom-built to find and exploit vulnerabilities in targeted systems.
“As threat actors increasingly move beyond basic prompting to deploying autonomous multi-agent frameworks, automated vulnerability exploitation pipelines, and alternative monetisation schemes… the adversarial AI landscape has reached a critical inflection point,” the company says.
Hackers are not only looking to steal financial details and other private information with such attacks, but also reportedly digital keys allowing access to a company’s AI usage allowances, with such stolen compute often then being sold on the black market.
Such “credential abuse”, which allows hackers to pretend to be authorised users of major companies for the purpose of accessing certain high-powered flagship models, was also a key focus of Microsoft’s research, which the company giving details of how it both detected and curtailed such attacks that touched its cloud services.
They added that working to detect such hostile cybercrime activity is getting more difficult, with hackers deploying increasingly complicated methods of obfuscating their actions.
Such attempts at hiding their AI-backed hacking attempts from authorities themselves reportedly leverage AI technology.
“Threat actors can and do change the source IP address, user agent strings, and other fingerprints left by the agentic AI penetration testing tools they use,” Microsoft says.
“This can make identifying and blocking malicious network traffic early difficult for defenders.”
Google’s research meanwhile cites Iran and North Korea as being two major state actors at the forefront of AI-powered hacking, with both countries reportedly leveraging Western systems to help their own cybercrime efforts.
The warnings come amid a spate of high-profile cyberattacks and global IT system outages, several of which have caused turmoil for the international economy.
Reports of an internal system used by the American National Security Agency (NSA) appeared to spark fears in the country’s most important stock markets, with the Dow Jones Industrial Average dropping 1,000 points during Tuesday trading.
Nevertheless, there has been no evidence presented that the outage had anything to do with a cybersecurity incident, with a similar outage plaguing air traffic communication across the United States also reportedly having been caused by an internal error as opposed to an external hack.
Other major incidents have been concretely traced to cybercrime, with North Korea said to have used artificial intelligence to help steal $6.3 billion worth of cryptocurrency from trading platform Binance earlier this month.
Closer to home, multiple credit unions were reportedly compromised as a result of a security vulnerability contained in software contracted from a third party. It is understood that some sensitive data was accessed during the incident, though it is unclear whether or not the hackers used artificial intelligence in any way.
Another incident earlier this week reportedly saw the websites of major firms such as the BBC, Air France, HSBC and Ikea compromised by hackers.
Cybercriminals were reportedly able to load malicious software onto over 200 major websites operated by trusted companies, with users accessing infected sites seeing malware automatically loaded onto their devices.
According to analysis of the attack, users who inserted any payment details onto the infected sites were liable to have their information stolen, while passwords and banking details loaded on a PC’s clipboard may have also been compromised.
As of writing, it appears that the vulnerability only impacted the immediately infected tab used to access the infected website, with closing the tab or restarting one’s computer said to be enough to purge the infection.
Hackers also reportedly used their access to the over 200 websites to infect any older iPhones accessing the websites. Analysts believe that the danger posed by this separate attempt at hacking users’ devices likely did not succeed, as Apple had already patched the exploited vulnerabilities several versions back.
As a result, it appears that few to no users using iPhones vulnerable to the hack accessed the impacted websites during the period when hackers had access.