The clinic’s office manager, Sarah, received an email appearing to come from Dr. Martinez (the clinic owner). The email referenced a real patient case they had discussed that morning, Sarah’s exact role, and a pending vendor invoice. It requested she approve a $3,400 wire transfer to a new account. The sender’s email address had one letter different from Dr. Martinez’s real address.

AI technique used:

Psychological tactic:

A veterinary technician named Carlos received a phone call from what sounded exactly like his father, saying he was in a car accident, needed bail money immediately, and asked Carlos to wire $1,800 to an account. Carlos almost complied before calling his father’s cell phone directly and confirming his father was safe at home.

AI technique used:

What would have happened without out-of-band verification?

The clinic received 847 emails in one week from different sender addresses, all with slightly different subject lines, all promoting the same fraudulent medical supplier. Traditional keyword filters blocked the first 200; the remaining 647 had been automatically rewritten by an AI tool to avoid the specific phrases that triggered the filter.

Which AI capability enabled this attack to evade filters?

This is an example of:

A security researcher found that an AI tool scanning for SQL injection vulnerabilities could also be used by attackers to automatically scan Maple Street’s patient portal for weaknesses. The same tool the clinic uses for defense was used offensively to map attack surfaces.

What does this demonstrate about AI security tools?

For any AI-based attack on the AP exam: identify (1) what AI capability was used to make the attack more effective, (2) what defense would have stopped it, and (3) whether human oversight was part of the solution. Business Email Compromise with AI personalization = out-of-band verification. Voice cloning = verify via known contact number. Filter evasion = add human review layer + model retraining.