Research paper

AI-Assisted Phishing Is Workflow Abuse With Better Output

The hard part is no longer producing a convincing email. The hard part is spotting where trust gets borrowed from brand, context, and habit before it reaches a human target.

This paper looks at how attackers use AI to compress the time and effort required for social engineering while still relying on classic trust mechanics.

Threat context

The persuasive layer gets cheaper, but the trust failure stays the same.

AI-assisted phishing changes the economics of deception. It lets attackers tailor language, style, and timing quickly, but the success path still depends on borrowed trust from vendors, managers, systems, or familiar workflows.

Technical analysis

AI helps attackers scale the parts of social engineering that are easiest to automate.

  1. Brand mimicry becomes cheaper and more convincing.
  2. Context can be rebuilt from public data and prior communications.
  3. The first human reviewer often sees a polished request instead of obvious malware.

The model is not the attack. The model just makes the old attack easier to ship.

Banjico working note

Defense moves

Defenses work best when they reduce the value of borrowed trust.

The defensive response should not depend on perfect human suspicion. Teams can add friction around money movement, identity changes, and sensitive requests so a polished message still has to pass a second control.

References