The Bait Got Better: How AI Is Changing Phishing

For years, security awareness training taught employees to look for the same warning signs: bad grammar, awkward phrasing, unusual greetings, and messages that simply didn’t sound right. Those clues existed largely because attackers faced practical limitations. They often didn’t speak the target’s language fluently and were sending the same generic message to thousands of people. Generative AI has largely removed those limitations.

Today, phishing emails can be exponentially more effective because they look polished, professional, and tailored to the industry and individual being targeted. They can reference a real project, vendor, colleague, or deadline and sound exactly like the type of message someone would expect to receive. Even the research has become easier. Public profiles, company announcements, conference appearances, and other publicly available information can be harvested from websites and social media, providing the context needed to quickly build a highly personalized message.

The Economics of Targeting Changed

When personalization becomes easier, the pool of worthwhile targets gets much larger. Detailed research on a controller at a mid-sized manufacturer was rarely worth an attacker’s time when that same hour could be spent sending a generic lure to a very large list. Automated personalization costs almost nothing per message, which means it is now worth doing for every controller at every manufacturer in a region.

Voice and Video Now Carry the Attack

Once an attacker has built a convincing pretext, there is no reason to keep it confined to email. Executives and business leaders often have publicly available recordings from conference panels, podcasts, webinars, interviews, and social media. That content can potentially be used to make impersonation attempts significantly more convincing. An email requesting an urgent transaction can now be reinforced by a phone call or video interaction designed to make the recipient believe they are communicating with someone they know and trust.

The Barrier to Entry Collapsed

The same technology making these attacks more convincing is also making them easier to carry out. Writing fluent business English in a second language, assembling an organizational chart from public sources, cloning a voice, and constructing a plausible pretext used to require either years of practice or a team with specialized skills. Much of that is no longer true. These capabilities are increasingly available by subscription, which means the population of people able to run a convincing targeted attack is far larger than it was and includes people with no technical background whatsoever.

Security Awareness Needs to Evolve, Too

If the message looks legitimate, the context is accurate, and the person on the phone sounds familiar, telling employees to simply “spot the phish” is no longer enough. Many security awareness programs still teach pattern recognition against a pattern that no longer exists and measure success by how many employees spot a simulated phishing email that was written to be spotted. The uncomfortable conclusion is that controls that depend on a person noticing something is wrong are now resting on exactly the judgment the attacker has engineered the message to defeat.

That makes the question less about whether an employee can recognize a convincing fake and more about what happens when they cannot. The controls that hold up best are the ones that never asked anyone to distinguish real from synthetic in the first place.

Will Your Controls Hold Up?

If you are unsure how your current defenses would perform against today’s more convincing attacks, our Risk Advisory professionals can help identify where the gaps are and where stronger safeguards may be needed.

In Part 2, “Don’t Take the Bait: Defending Against AI-Driven Phishing,” find out which controls and processes can help organizations strengthen their defenses as phishing becomes more convincing and harder to recognize.