AI is now attacking AI, and we learned this the hard way.

A little while back we got hit with 600 calls in under an hour. $1,000 in usage fees. It was just a recording playing over and over again.

This was our first time dealing with something like this. It came out of nowhere and caught us completely off guard. By the time we realized what was happening, the damage was done.

I've seen chat-based AI systems get spammed before, but phone attacks? Those cost real money to execute at scale... unless you own the phone lines. 🫠

So what would you have done?

I spent a lot of time thinking through this. What would a real call center or business do if their phones suddenly started ringing nonstop with the same recorded message?

They'd catch on pretty quick.

I extracted the main elements real people use to solve these problems:

1. Call centers & businesses know their normal volume. Calls per day, per hour, per minute, busiest times. We needed to track these metrics too and build a system that could use them to detect attacks.

2. Pattern recognition. If you picked up just TWO calls that sounded identical, even from different numbers, you'd know something was up and hang up immediately.

The solution: A live supervision system

We built an extremely low-cost monitoring layer that watches active calls in real time. An Algorithm + LLM analyzes:

Call duration and how long it's been going

Time between turns (if there's 30+ seconds of silence over just 2 turns, that's a red flag)

Interruptions (how often the caller interrupts, which helps identify recordings)

Recent global activity for that number, plus a snapshot of the last hour (example: "calls in last hour: 334, average for this hour: 34, pattern:

nonsensical caller asking why sneezing makes the world rotate differently")

Plus other safeguards like progressive throttling if the same number calls repeatedly, with automatic cutoffs after a threshold. This can apply to area codes and a lot of the times the first 3 digits of the number are also the same which helps you pick up and stop these attacks earlier on.

We went from vulnerable to protected by building a system that thinks like a human call center.

I hope this helps anyone who's implementing a voice AI systems in their business.

9 comments

AI is now attacking AI, and we learned this the hard way.