Phishing emails about new AI tools got clicks from 45% of recipients in Kaspersky's latest test data, the highest of any theme the company measured.
The figures come from anonymized data that customers voluntarily provided from Kaspersky's Phishing Simulator, a tool that sends fake lures to staff. Five topics performed best: AI and tech, healthcare and insurance, IT policies and compliance, gifts and promotions, and HR and corporate communications. Healthcare lures averaged 28%, IT policy 26%, gifts 24% and HR 22%. The standouts were individual emails. One about a sanitary inspection report reached 92%, and an invitation to wear a national football jersey for a prize drew 66%.
Those numbers mix category averages with single-lure results, so they work better as a pattern than a ranking. The pattern is that specificity beats topic. A message that matches an employee's workflow or locale gets opened without suspicion, which Kaspersky's Svetlana Kalashnikova said will accelerate with generative AI.
That shift matters because the old tell is gone. Clumsy phrasing from non-native writers used to give phishing away, and AI-written text now reads like a real communications team. Kaspersky says detection has moved to the links, attachments and sender domain. Urgency such as a 24-hour suspension threat, unfamiliar third-party document hosting and executable attachments remain red flags. So does a mismatched sender, such as the UK postal service writing from a Proton address.
Kaspersky also claims regular training and simulations cut susceptibility from 60% to 5-7%, citing its own KASAP analytics. A vendor that sells awareness training has a stake in that number, and the source gives no detail on how it was measured.
The more useful question is what training should teach once polished writing is no longer a clue. Checking where a message came from and where its links lead is a habit, and the 92% result suggests it needs to hold up against emails that look like a normal Tuesday task.