AI Bots Are Turning Scammers’ Own Tactics Against Them
Lead
AI is being repurposed from a tool that enables scams to a shield that ties up scammers for hours. Companies like Australia’s Apate are deploying hundreds of thousands of conversational bots that answer scam calls and texts, while researchers are adding large language models (LLMs) to honeypots to make them look more real.
How Apate’s Bot Swarm Works
Apite’s platform runs about 350,000 bots that answer inbound scam calls and infiltrate scam‑related chat groups. Each bot pretends to be a real person – it can have a WhatsApp profile, a name, a language preference, or even a “busy” excuse. The goal is to keep the scammer on the line long enough to waste the time they could have spent targeting genuine victims. While the bot talks, it logs details such as the scam URL, money‑mule accounts, and any bank information the scammer mentions. To date the system has harvested more than 250,000 pieces of information in real time.
Why Bots Beat Traditional Call‑Blocking
Traditional call‑blocking services simply hang up on suspicious numbers. That stops the call, but it also gives the scammer a clean failure signal and no insight into their tactics. Apate’s bots, by contrast, simulate a genuine, if skeptical, conversation. In testing, a single call often lasted over two hours before the bot politely ended the interaction. The longer a scammer stays on a call, the fewer fresh numbers they can reach, reducing the overall volume of potential victims.
LLM‑Powered Honeypots: Making Traps More Convincing
Honeypots are deliberately vulnerable computers or services set up to attract attackers. Historically they behaved in predictable ways, which savvy attackers could spot quickly. Researchers at ETH Zurich added an LLM to a honeypot’s response engine. Their experiments showed that AI‑driven honeypots kept attacking agents engaged significantly longer than static traps, and the agents were less likely to flag the system as a honeypot. The LLM can generate plausible system messages, mimic human error, and adapt its replies based on the attacker’s behavior, making the deception harder to break.
The Trade‑off No One Talks About
AI‑Enabled Distraction vs. Resource Drain
Deploying hundreds of thousands of bots and LLM‑enhanced honeypots sounds like a win, but it also consumes bandwidth, compute cycles, and human oversight. Each bot must be monitored for accidental leaks of real personal data, and the LLMs require regular updates to avoid being trained on malicious content. Moreover, scammers can adapt – they may start using voice‑synthesis verification or pattern‑recognition tools to spot bot‑like speech. The real cost, therefore, is not just the infrastructure but the ongoing arms race to keep the bots believable.
What to Watch Next
- Regulatory responses – Governments may impose limits on automated call answering to protect privacy, which could force bot operators to obtain explicit consent.
- Scammer counter‑measures – Early reports suggest some fraud rings are testing AI voice‑cloning to detect synthetic voices; watch for any decline in bot engagement times.
- Cross‑industry data sharing – The article notes that better intelligence sharing between banks, telecoms, and law enforcement could amplify the bots’ impact. Look for pilots that integrate bot‑collected intel into real‑time fraud‑prevention feeds.
Practical Steps for Individuals and Small Businesses
- Enable carrier‑level spam filtering – Even if you don’t use a full‑scale bot service, most carriers now offer AI‑assisted spam detection that can divert suspicious calls.
- Report suspicious numbers – Forward scam calls or texts to your local anti‑fraud hotline; the data helps train larger bot networks.
- Educate staff – Teach employees to recognize the hallmarks of a bot conversation (overly consistent politeness, scripted skepticism) so they don’t waste time on false positives.
- Consider a lightweight honeypot – Small firms can deploy a simple email address that auto‑replies with AI‑generated messages to lure phishing attempts and capture malicious payloads.
