Feds Warn Healthcare Sector of AI-Augmented Phishing ThreatsGen AI Helps Hackers Create More Realistic Phishing Messages to Infiltrate Systems
Hospitals, clinics and doctor practices have long fallen victim to cyberattacks and breaches kicked off with phishing emails. But with the advent of AI-augmented phishing, the lures are more convincing and could lead to even more scams targeting healthcare organizations, federal authorities warned.
The Department of Health and Human Services' Health Sector Cybersecurity Coordination Center in an advisory issued Thursday warned healthcare sector organizations to prepare for the growing threats posed by AI-augmented phishing.
Phishing is a common and lucrative tactic used by hackers to trick users into sharing credentials, downloading malware including ransomware, and stealing the sensitive data of healthcare organizations, HHS HC3 noted.
"The advent of artificial intelligence has only made phishing attempts more effective, especially since those tools are freely available to the public," the agency warned.
Federal officials and other experts warn that generative AI tools can create more realistic spear-phishing messages from senior leaders to lower-level employees. For example, OpenAIs ChatGPT enables users to choose personas of management roles such as CEO, CFO or CMO and emulate their style of writing.
"The ability for gen-AI to emulate the grammatical style and persona of a person in management such as the CEO, CMO and others will very likely lead to a new uptick in AI-assisted phishing as a preferred initial access vector," said Mike Hamilton, CISO and co-founder of security firm Critical Insight.
"This preys on one of the cognitive biases we all have - authority bias, or being unconsciously influenced by someone in a position of authority," Hamilton said, adding that AI adoption will "very likely lead to a new uptick in AI-assisted phishing as a preferred initial access vector."
As ChatGPT and other generative AI tools fall into the hands of cybercriminals, phishing attacks are getting more difficult to detect.
"Now, even inexperienced and non-technical threat actors can launch sophisticated phishing emails," said Mike Britton, CISO at security firm Abnormal Security.
"AI-generated email attacks are the most concerning AI cyber threat today because they are such an easy and effective tactic for cybercriminals," said Britton, adding that "we’re already seeing incidents of these kinds of attacks."
In fact, recent research from Abnormal Security found that 80% of security leaders confirmed being hit by AI-generated email attacks - or at least strongly suspected the involvement of AI tools, he said.
The healthcare industry is particularly susceptible to phishing attacks because of the massive volume of attacks targeting this industry compared to others, Britton said.
Also, the healthcare sector has a vast supply chain and partner network, giving threat actors an opportunity to use phishing to distribute fake insurance claims or fake invoices for medical equipment, which can appear to be a normal part of doing business to healthcare workers and administrative staff, he said.
HHS HC3 also confirmed that generative AI tools are already being used by attackers, citing FraudGPT, developed for bad actors to craft malware and text for phishing emails. "It is available on the dark web and on Telegram for a relatively cheap price – a $200 per month or $1,700 per year subscription fee - which makes it well within the price range of even moderately sophisticated cybercriminals," HHS HC3 said.
Prevention to falling victim to all phishing attacks, including those augmented by AI, takes a defense-in-depth approach and vigilance, HHS HC3 advised.
That includes configuring email servers to filter spam emails or implementing a spam gateway filter platform, using endpoint security software and multifactor authentication. Another important factor is workforce awareness, HHS HC3 said.
"Staff training should include examples of AI-generated phishing, along with a discussion of those cognitive biases that will be exploited," Hamilton suggested.
In addition, "limiting access to the internet - including email - for staff that do not need external access and enforcing a policy of personal use on a personal device will go furthest in closing off this attack vector," he said.
In the near term, "email filtering will begin assessing every message for AI-generated content, external domain sources and other indicators -however, these products are not yet generally available," he said.
In any case, the methods of an individual detecting an AI-generated phishing message is not much different than the process of detecting “traditional” phishing, he said.
"However, the combination of better grammar, persona emulation and cognitive biases make this detection regimen less likely to occur. Assistance from technology is needed to perform the identification or to bar users from internet-sourced messaging," he said.
While the healthcare sector is certainly still prone to phishing, in part because of the urgency demands on clinical and other staff, the constant and repeated training that many healthcare entities now routinely provide their workforces seems to be helping, Hamilton said.
While healthcare sector firms need to be proactive about defending against emerging AI-augmented phishing schemes, they should also be mindful of threats involving AI deployments within their own environments, Hamilton said.
"Internal use of AI to improve staff efficiencies and Internet-facing AI assistants can also be misused, if not carefully trained, to provide information about internal documents, architectures, employees and processes," he said.
This information can be used to discover and develop attack and compromise strategies, he said. "AI for internal use should be carefully developed along with policies that take the limitation of training information into consideration."
Britton also suggests that healthcare organization need to be mindful of other AI-generated threats.
"These may include polymorphic malware, where generative AI could be used to automatically alter its source code and behavior to evade detection, or increased endpoint exploits, where attacks leverage generative AI to customize attacks to known vulnerabilities," he said.