AI Is Changing the Speed at Which Fraud Evolves

Visa detected around $1 billion in scam transactions in the second half of 2025 alone. Scams are now the largest category of consumer payment fraud.

Another report found that AI-enabled scams increased losses by almost 20% year over year, reaching $14.3 billion globally.

The obvious conclusion is that AI is making fraud worse.

I think there's a more interesting conclusion.

AI is changing the speed of fraud evolution

Fraudsters have always adapted. The difference is that adaptation used to take time. A new fraud act gets implemented, fraudsters test it, they find a weakness, and adjust their behaviour.

Today, that entire cycle happens much faster. They can generate phishing campaigns at scale, create convincing fake identities, test hundreds of variations, learn from the results, and launch the next attack almost immediately.

The challenge for merchants is that fraud is becoming more adaptive. That changes how fraud systems need to operate.

• A fraud strategy that gets reviewed every few months isn't enough
• Fraud acts can't be written once and forgotten
• Models can't be treated as permanent solutions

The question becomes: can our fraud strategy evolve as quickly as the fraud itself?

That means continuously reviewing fraud patterns. Understanding which behaviours have changed, removing outdated acts, and focusing on signals that are harder to manipulate than individual data points.

The fundamentals still matter

The fundamentals of good fraud management remain the same:

1. Understand the behaviour
2. Ignore the noise
3. Build systems that evolve as quickly as the threats they're designed to stop

If you want to see how FraudPulse turns shifting transaction patterns into ranked rules and actions, book a walkthrough.

Originally shared on LinkedIn.