AI Phishing Raises the Bar for Employee Training and Incident Response

Phishing readiness is bigger than spotting a suspicious link

Phishing simulations can show whether employees recognize a familiar lure. They say less about how an organization will handle a convincing message that looks like ordinary business. My read is that security teams need to test both: whether people verify unusual requests and whether the organization can act quickly when a suspicious message gets through.

That shift matters as generative AI helps attackers write more polished messages, personalize lures and draw on relevant business context. It doesn’t mean every campaign uses AI. It does mean employees may encounter fewer of the obvious clues that awareness programs have traditionally emphasized, such as spelling errors, generic greetings and awkward formatting.

A 2025 report by SC Media, covering research by Egress, found that 75% of the phishing kits examined on dark-web markets advertised AI capabilities and 82% offered deepfake features. Those figures describe the kits analyzed, not phishing activity as a whole. Still, they point to AI features appearing in criminal tools. Recent phishing attacks involving AI prompt injections add another reason for teams to revisit what their exercises measure.

Measure whether people report suspicious messages

Click rates are easy to track and present to senior management. Simulations can also measure whether recipients enter credentials, delete a message or report it. But identical click rates can hide very different outcomes: one group might quickly flag a suspicious email, while another quietly deletes it. In the first case, analysts have a chance to investigate, find other recipients and warn them.

Reporting rates and time to report show whether employees use the approved channel and how quickly a concern reaches the security team. Teams should also check whether reports lead to effective action. The National Cyber Security Centre’s phishing guidance offers practical advice on identifying and reporting suspicious messages; organizations can make such guidance available through their websites or HR portals.

Simulation results still need context. A controlled exercise can reveal weaknesses, but it can’t guarantee how someone will react during a real attack. A recipient’s circumstances may affect their response, and repeated campaigns can teach people to recognize familiar templates. Clicking a simulated link also doesn’t, by itself, show that the person would expose the organization during a real incident.

Use the results to improve defenses, not to rank employees by pass or fail. The way teams communicate a mistake matters: staff should be able to report a suspicious message without fearing embarrassment or punishment.

Exercise the response, not just the inbox

Scenarios should resemble the requests employees actually handle. That could mean an unexpected document share, a supplier asking for a change or a message impersonating a trusted colleague. The exercise should ask whether recipients verify unusual requests through a trusted channel and report concerns, rather than focusing only on links and credential forms.

Building those exercises may require custom mailboxes, different simulation procedures and several levels of approval. That work connects awareness training to the steps employees need to take when a message feels wrong.

Security teams should also test what happens after a report arrives. Can analysts identify other recipients and contact them? Can the organization remove the message from inboxes, block malicious infrastructure and investigate possible account compromise? Those capabilities show whether training supports incident response in practice.

Exercises should include security, IT, finance and the relevant business teams, so people know their roles when an attack succeeds. No program will stop every click. Faster reporting and coordinated response can still limit the damage.

In simulations I’ve conducted, including one in October 2026, the useful question isn’t simply whether employees recognized a known pattern. It’s whether they know how to check a plausible request and whether their organization can detect and contain an attack. Yesterday’s tests can offer lessons, but tomorrow’s readiness depends on what happens when the warning signs are harder to see.