Apate Sends AI Agents Deep Undercover to Get Scammed and Gather Fraud Intelligence

While most companies are busy using AI agents to speed up product development or optimize workflows, a startup called Apate.ai is doing the exact opposite — actively sending AI agents out to get scammed.

Their goal isn't to close a deal, but to be defrauded. Every scam is an intelligence-gathering mission.

Using AI Agents to Play the Victim

Apate.ai's core idea is counterintuitive: rather than passively waiting for fraud to happen and then scrambling to fix things, it proactively places AI agents into scammers' line of sight, having them pose as potential victims and gather intelligence on fraud tactics, script patterns, payment channels, and infrastructure through interactions with fraudsters.

These AI agents aren't simple chatbots. They're designed to mimic real users' conversational rhythm, hesitation, skepticism, and even the process of "taking the bait," inducing scammers to reveal more operational details.

Why "Getting Scammed" Is an Intelligence Capability

Traditional anti-fraud relies mainly on two approaches: post-mortem analysis of fraud cases that have already occurred, or real-time interception based on rules and models. Both have obvious shortcomings — the former is always a step behind, while the latter is easily bypassed by scammers' rapidly iterating scripts.

Apate's model offers a new dimension: proactive fraud intelligence. Through real-time interaction with AI agents, enterprises can obtain first-hand information about fraud campaigns, including:

• Currently prevalent fraud scripts and script structures
• Payment channels and money flow methods used by scammers
• Characteristics of fraud infrastructure (domains, accounts, communication tools)
• Differences in targeting patterns across regions and demographics

This intelligence can be fed directly back into anti-fraud systems, forming a closed loop of "collection — analysis — defense."

Proactive Fraud Intelligence is a new paradigm that has emerged in the anti-fraud field in recent years. Traditional fraud intelligence mostly comes from post-incident reports, monitoring of underground forums, or purchasing third-party data, which suffers from lag and coverage blind spots. Proactive intelligence, by contrast, deploys simulated agents, honeypot accounts, or disguised identities to interact directly with scammers, capturing their scripts, tools, and money trails before or during an attack. This approach has precedent in cybersecurity — for example, honeypot technology disguises vulnerabilities to attract attackers and record their methods. Apate.ai extends this logic from the network layer to the social engineering layer, replacing static honeypots with conversational AI agents that can handle more complex voice, text, and multi-turn interaction scams. Its core value lies in shortening the intelligence acquisition cycle and shifting defenders from "guessing the attacker's next move" to "observing the attacker's current behavior."

How It Differs from Existing Anti-Fraud Solutions

There are already many AI-driven anti-fraud products on the market, but most focus on detection and interception. Apate's differentiation lies in intelligence generation, not just intelligence consumption.

This also means its value may not lie in replacing existing risk control systems, but rather in serving as an upstream intelligence source, providing banks, e-commerce platforms, payment institutions, and others with more timely data that's closer to real attack methods.

From a business perspective, this model faces very practical challenges: how to operate large numbers of AI agents at scale, how to ensure agent behavior doesn't cross legal and ethical boundaries, and how to transform unstructured conversational intelligence into actionable risk control rules.

AI-driven anti-fraud products typically rely on supervised learning or anomaly detection models to score and intercept structured data such as transactions, logins, and device fingerprints in real time. These systems depend on historical labeled data and respond slowly to new or rapidly mutating fraud tactics. Apate.ai, by contrast, generates unstructured conversational intelligence that requires entity extraction, intent recognition, and knowledge graph construction before it can be converted into risk control rules or features. The two aren't substitutes: the former is the "execution layer" responsible for immediate blocking; the latter is the "intelligence layer" responsible for discovering unknown threats and updating defense strategies. This layered architecture is increasingly common in financial risk control — for example, banks separate threat intelligence platforms from real-time decision engines, with the former continuously collecting and the latter responding quickly. Apate's challenge lies in how to automatically map vague expressions in conversations (such as "transfer to a certain account") into machine-readable indicators (IOCs), while ensuring the timeliness and accuracy of the intelligence.

Outlook and Concerns

Using AI to fight AI fraud sounds like an inevitable trend. But the path of "fighting deception with deception" also raises noteworthy questions. AI agents may come into contact with real people during the scamming process (if the scammer is also a real person) — does their behavior constitute some form of entrapment? How is the collected intelligence used, stored, and shared?

Apate.ai is still in its early stages, with limited public information. But its approach points in a direction: as AI agent capabilities rapidly advance, the battlefield of anti-fraud is shifting from "defense" to "infiltration."

For security teams, this may be a new category worth tracking.

Entrapment in legal terms typically refers to law enforcement or agents inducing someone with no prior criminal intent to commit a crime. Apate.ai's AI agents actively interacting with scammers may raise similar controversies: if the agent's responses are too "cooperative," does that constitute encouragement or assistance of fraudulent behavior? In most jurisdictions, the entrapment defense applies mainly to criminal proceedings and requires proving the defendant had no prior criminal intent. Scammers who proactively contact the agent and carry out fraud typically already have criminal intent, so entrapment risk is low. But civil and privacy issues are more complex: must the agent disclose its nature to real-person scammers? How do collected voice, text, and payment information comply with data protection regulations like GDPR and CCPA? Additionally, if the agent actually makes payments during interactions, it could involve money laundering or legal obstacles to fund recovery. These boundaries need to be defined jointly by enterprises, regulators, and ethicists.

One-line Summary

Apate.ai has AI agents actively pose as victims to gather fraud intelligence from interactions with scammers, opening a new path for proactive anti-fraud.

Paragraph Summary

Apate.ai is a startup that goes against the grain, sending AI agents to actively interact with scammers and pose as potential victims to collect fraud intelligence. This "proactive fraud intelligence" model differs from traditional post-mortem analysis and real-time interception, obtaining first-hand data on fraud scripts, payment channels, and infrastructure to form a collection-analysis-defense loop. Its value lies in serving as an upstream intelligence source, providing banks, e-commerce, and payment institutions with more timely data on attack methods. However, the model also faces challenges in scaling operations, legal and ethical boundaries, and converting unstructured intelligence. The company is still early-stage but represents a new direction for anti-fraud shifting from defense to infiltration.