Apate.ai uses 350k AI decoys to waste scammers' time; bank scam losses down 76%
Core Highlights
The author tells two cases of using AI to fight back against scams. Australian company Apate.ai uses 350,000 AIs disguised as gullible victims, having chatted with scammers 2.5 million times and wasting about 17 years of scammers' time. The Commonwealth Bank of Australia's publicly reported scam losses dropped 76% as a result. This is a real-world sample of using AI against AI scams, turning anti-fraud from passive blocking into active attrition that exhausts the attacker's resources, and it is the first large public data point on the defense side winning.
What Happened
Apate.ai's logic is clever. It does not block texts or ban accounts, but sends a swarm of AI decoys to keep scammers talking, wasting their time in conversations that will never close. Australian carrier TPG routed nearly 590,000 suspicious calls to these AIs, pouring sand into the very front of the scammer funnel and diluting conversion efficiency at the source. The scale matters. Hundreds of thousands of decoys mean scammers burn their most scarce resource, human attention, on targets that will never pay, which is the resource that actually limits a scam operation's throughput.
Technical Details
To seem like a real person, such systems need speech synthesis, dialogue strategy, emotion recognition, and boundary control of biting but never transferring money. The hard part is not chatting fluently, but making scammers believe they are dealing with a real victim without triggering risk controls that end the conversation. Apate.ai raised an $8.15M seed round this August, signaling capital backs the approach, which means defense as attrition war is being productized into a service rather than remaining a research curiosity that never reaches production.
Compared With Alternatives
Traditional anti-fraud relies on interception plus user education, passive and lagging. Apate.ai actively consumes scammer resources, shifting defense from blocking to draining. Young people in China using AI video to counter AI sob-story accounts are a grassroots version of the same idea, showing the confrontation is moving from single points to scaled automation, and that the same generative tools empower both the attacker and the defender.
Industry Impact and Use Cases
For carriers and banks, these AI decoys can serve as a new layer on the anti-fraud front line, sitting upstream of the human who would otherwise be targeted. For society, it signals a trend that the contest between scams and anti-fraud is going fully AI-native, and whoever has larger automation scale and higher human-likeness gains the upper hand. Future anti-fraud may evolve into a protracted war of AI armies against AI armies.
Practical Notes
The insight for defenders is to attack the scammer's economics rather than only the victim's behavior. A call that wastes twenty minutes of a scammer's time is worth more than a hundred blocked messages, because the scarce resource is the human on the other end. Deploy decoys at the funnel's top, where suspicious traffic is densest, and measure success in scammer-minutes wasted, not messages blocked.
What to Watch
Watch whether regulators accept such active deception, since trapping scammers walks a fine line between defense and entrapment in some jurisdictions. The bigger signal is that fraud defense is becoming a compute-and-conversation arms race, and incumbents that stay passive will lose share to AI-native startups that can field believable automated conversations at scale.
Takeaway
Apate.ai shows the next front in cyber-defense is not a wall but a swarm, and the side that can field more believable automated conversations wins the war of attrition, a pattern that will repeat across every domain where humans are the bottleneck.
Background
Apate.ai sits at the intersection of two trends that are accelerating together, scams are scaling with generative AI, and defense is scaling with generative AI in response, so the two are locked in an arms race where each side's automation feeds the other's. The 76% loss reduction at Commonwealth Bank is the first large, public data point that the defense side can win, and it reframes anti-fraud as an offensive as well as a defensive problem.
Where to Look Next
Watch regulatory reaction, because active deception of scammers is legally grey in some jurisdictions, and how regulators classify it will shape whether this becomes a mainstream defensive layer or stays a clever experiment confined to a few carriers willing to take the risk of entrapment claims.
Closing Thoughts
The takeaway is that the future of fraud defense is a swarm against a swarm, and the side that can field more believable automated conversations at lower cost will win, which means the defenders need their own AI factories as much as the attackers do.
Final Note
The uncomfortable truth behind the 76% number is that defense at scale is now a production engineering problem, not a policy problem, and the carriers who win will be the ones who treat scammer-attrition as a system to be optimized rather than a campaign to be run. That means measuring scammer-minutes wasted, tuning dialogue strategies like any other funnel, and accepting that the fight is perpetual. The alternative, passive blocking, is losing to an opponent that never sleeps and never stops iterating.