Phishing tests lag behind AI lures

Most phishing simulations still measure whether an employee clicks a link or enters a password on a lure that looks like a decade-old template. That test no longer matches the messages workers are likely to face, according to a practitioner who has run multiple simulations, including one in October 2026. The gap between what organizations drill and what attackers now produce is the central problem: training people to recognize warning signs that AI can smooth away leaves them exposed to the ones that remain.

The argument is not that phishing simulations are worthless. It is that they are being pointed at yesterday's threat while the tools available to criminals have moved on. Fat Joe's 2021 line — “Yesterday’s price is not today’s price” — opens the source piece, and the comparison to security awareness is deliberate: what worked previously may not deliver value now.

Signs that AI erases

Traditional phishing tells are getting less dependable. Generative AI lets attackers produce convincing messages, personalize lures, and raise the quality of their communication. That does not mean every campaign uses AI, but it raises the bar for what employees may encounter.

The tools themselves reflect the shift. A 2025 report by SC Media covering Egress research found that 75% of the phishing kits examined on dark-web markets advertised AI capabilities, while 82% offered deepfake features. Those numbers describe the kits analyzed, not all phishing activity, but they show AI-enabled features showing up in criminal tooling.

Separately, there has been a spike in recent phishing attacks with AI prompt injections. Taken together, the picture that emerges is of a landscape that has morphed well beyond the misspelled email with a generic greeting.

Why recognition training falls short

If employees are mainly taught to look for spelling mistakes, generic greetings, or suspicious formatting, they may be less prepared for a message that uses familiar names, relevant business context, and professional language — all things AI can supply. The warning signs the training emphasized simply do not apply to a well-crafted lure.

The practical response, the source argues, is to shift emphasis toward verification rather than pure recognition. Unexpected requests should be checked through a trusted channel, even when the message appears legitimate. That means teaching people a process they can run when their instincts fail, not just a checklist of red flags to memorize.

Organizations looking for grounding material can point staff to the National Cyber Security Centre’s guidance on phishing, which offers practical advice on recognizing and reporting suspicious messages. The source suggests that guidance should be available on an organization's website or HR portals rather than buried in a training module.

Click rates hide the real picture

Simulations tend to produce a tidy headline: the percentage of recipients who clicked a link, entered credentials, and deleted or reported the message. Those numbers are easy to communicate, compare over time, and present to senior management. They also capture only one part of the story.

Consider two groups with identical click rates. In the first, employees who suspect a message report it quickly, letting the security team investigate, find other recipients, and warn them. In the second, people delete the message or ignore it without reporting. The click rates look the same, but the organization's ability to detect a real campaign is not.

Reporting rates and time to report deserve more attention than they typically get. Security teams should know whether employees use the approved reporting channel, how quickly suspicious messages reach analysts, and whether those reports lead to effective action. A fast report can turn one compromised inbox into a contained incident; silence can let a campaign spread unnoticed.

Interpreting results with care

Simulation results need careful reading. A test is a controlled exercise, not a guarantee of how someone will behave during a real attack, which may depend on the recipient's state of mind. Employees can also learn to recognize familiar templates or memorize the patterns used in repeated campaigns, which inflates pass rates without improving real-world judgment.

Clicking a simulated link does not automatically mean someone would compromise the organization in a real incident. The source notes that this distinction matters for how results are communicated — the objective should be to identify weaknesses and improve security, not to rank employees by how often they pass or fail. That principle should carry through to the way messages about failure are delivered.

Shame and punishment are poor teaching tools for reporting. If an employee fears embarrassment for clicking, the next suspicious message they receive is more likely to be deleted quietly than reported through the channel the security team depends on.

Scenarios that match the job

Simulations should evolve alongside the threats they represent. Scenarios should reflect realistic business situations — unexpected document sharing, supplier requests, and attempts to impersonate trusted colleagues — instead of relying only on clicking links or supplying credentials.

Employees should learn how to verify unusual requests and report suspected attacks without fear of embarrassment or punishment. From experience, the source notes, this can involve many moving parts: custom mailboxes, a departure from the simulation norm, and multiple levels of approval before exercises can be run. It is still worth the effort.

The point is not to make drills harder for their own sake. It is to make them resemble the messages that will actually arrive, so that the skills being tested are the ones that will matter.

Measuring the response, not just the person

Security teams should also evaluate what happens after a suspicious message arrives. How quickly is it reported? Can analysts identify and contact other recipients who may have responded? Can the organization remove the message from inboxes, block malicious infrastructure, and investigate potential account compromise?

Those questions connect awareness testing to operational readiness. A simulation that measures only individual behavior tells you nothing about whether the security team can act on a report once it lands.

The same principle applies to incident exercises. Security, IT, finance, and relevant business teams should know what to do when a real phishing attack succeeds — which, the source notes, it inevitably will, someday. A program that improves reporting and response can deliver value even when it cannot stop every employee from clicking.

The data behind the shift

The numbers cited in the source piece are narrow but pointed — they describe the criminal tooling market and the features advertised within it:

• 75% of phishing kits examined on dark-web markets advertised AI capabilities, per a 2025 report by SC Media covering Egress research.
• 82% of those kits offered deepfake features.
• One phishing simulation referenced by the author took place in October 2026.

Those figures relate to the kits analyzed rather than all phishing activity. The source is explicit about that limit, and it is worth keeping in mind: the numbers show AI features appearing in tools available to criminals, not that every attack uses them.

What to do with yesterday's lessons

AI will keep changing how phishing messages are created and delivered, and security awareness has to change with it. The question the source poses is whether an organization can verify suspicious activity, detect an attack, and limit its impact when the message looks entirely legitimate and is prone to being clicked anyway.

Yesterday's tests can still teach useful lessons. The argument is that they should not define tomorrow's readiness — that awareness programs built around recognizing familiar errors need to add verification steps, measure reporting behavior, and rehearse the response that follows a click.

Why it matters

For organizations running awareness programs, the implication is that a clean simulation dashboard may say less than it appears to. If the drills are built on warning signs AI can erase, a low click rate could reflect employees who have learned the test rather than employees who can spot a real lure. This suggests security teams should weigh reporting speed and the quality of follow-up alongside click percentages when judging whether a program works.

The source does not claim that simulations are obsolete, only that their focus needs to move. Organizations that treat awareness testing as a checkbox may find themselves measuring the wrong thing — while attackers build messages that pass inspection precisely because they carry none of the defects employees were trained to catch.