A study by IBM’s X-Force security team has shown that AI is getting better at online deceptions like phishing. This conclusion was reached after a thorough experiment.

IBM X-Force, the American tech giant’s security team, conducted an experiment to find out how good AI is at online deception. The researchers examined how technology compares to human efforts at phishing.

You can learn to experiment

New technologies almost always mean that there are people who can use them for less positive things. Of course, this also applies to generative AI, because cybercriminals have long since discovered how to use artificial intelligence. IBM X-Force examined the advancement of these capabilities.

To find out, the team investigated whether generative AI can think as deceptively as the human mind. IBM X-Force turned it into a phishing competition.

Work

The researchers managed to get a generative AI model from ChatGPT to create a highly successful phishing email in just five prompts. What’s even more impressive is that the system took just as many minutes to do this. To put that in perspective, IBM X-Force members typically spend around sixteen hours working on an ideal phishing email.

The model was asked to create emails for specific industries. The researchers then had the system identify relevant areas of focus for workers in each sector. Finally, the AI ​​had to develop social and marketing techniques to optimize the emails.

This resulted in an email aimed at healthcare workers (a popular sector for scammers) with information about jobs and career opportunities. The text had to be personal, while at the same time exuding confidence and authority and calling for action. Furthermore, the message had to come from HR, so to speak.

The result was impressive:

Human edge

Ultimately, after writing and sending an email, the human team was able to convince a few more people, but there wasn’t much room for maneuver. The human approach was only three percent more efficient than AI.

The conclusion is that human EQ and personalization work even better for now. An important detail was that the AI ​​had also chosen an email subject that was far too long.

AI is therefore currently far from being the perfect tool that criminals can work with. However, X-Force emphasizes that technology is evolving quickly and constantly. Therefore, companies and individuals must also learn to think ahead, and caution is of course always necessary. Despite the many warnings and incidents, phishing remains a popular tactic.