Artificial intelligence is no longer just a tool for stopping fraud. New research shows it may now be better at committing it.
A joint study from Northeastern University, Ariel University, and the research arm of cybersecurity firm Gen Digital has found that AI chatbots can be more persuasive than human scammers when running a common type of cyberattack known as “pig butchering.” The findings arrive as security professionals across the world are already bracing for a new generation of AI-enabled cybercrime.
The research adds fresh urgency to a debate that has circulated in cybersecurity circles for years: as generative AI tools become cheaper and more accessible, the same technology defending networks from a cyberattack could just as easily be repurposed to launch one. For everyday internet users, the implications are direct. The tell-tale signs people have relied on for years, such as awkward phrasing or robotic replies, may no longer reliably separate a scam from a genuine conversation.
What the Study Found #
Researchers set out to answer a narrow but important question: when it comes to executing a real-world social engineering scheme, does an AI chatbot actually perform better than a human con artist? To test it, they built a simulated version of a pig butchering scam, a scheme in which a fraudster spends time building a relationship with a target before steering them toward a fraudulent app, investment platform, or fake financial opportunity.
AI Chatbot Outperformed Human Scammer
In head-to-head testing, the AI-driven chatbot proved more effective than its human counterpart at convincing participants to download a piece of software designed to mimic a helpful tool. In reality, the simulated app was built to harvest sensitive data such as login credentials, mirroring how a genuine cyberattack often unfolds once trust has been established.
According to reporting from Vice, which reviewed the study, participants placed a notably higher degree of trust in the AI-generated conversations compared to the ones run by a human operator. Several participants only realized they had been manipulated after they had already handed over their credentials.
Why AI May Be a More Dangerous Cyberattack Tool
Researchers point to several structural advantages that make AI systems particularly effective at running a large-scale cyberattack campaign:
| Advantage | Why It Matters |
|---|---|
| Scalability | AI can chat with thousands of potential victims at once, something no individual scammer can match |
| Consistency | Chatbots maintain the same tone and story across every conversation without slipping up |
| Instant response | Victims get replies immediately, keeping engagement high and suspicion low |
| Adaptive messaging | The AI adjusts its approach in real time based on how a target reacts |
| Round-the-clock availability | Unlike human criminals, AI systems do not need breaks, shifts, or sleep |
These traits, researchers say, allow a single AI-powered operation to run a cyberattack campaign at a scale that would previously have required an entire team of human operators.
The Double-Edged Role of AI in Cybersecurity #
Same Technology, Opposite Uses
Law enforcement agencies, banks, and technology companies already lean heavily on machine learning to flag suspicious transactions, detect malware, and block a cyberattack before it reaches a user. This study underscores a growing tension in the industry: the large language models built to protect people are structurally similar to the ones that can be fine-tuned to exploit them.
Gen Digital, whose research team contributed to the study, has previously highlighted how consumer-facing scams are evolving alongside AI capabilities, noting that fraud detection tools must adapt at the same pace as the threats they are designed to catch.
A Warning for Individuals and Businesses
The researchers frame their findings as a call to action rather than a cause for panic. As AI-generated conversation becomes harder to distinguish from a real human, the traditional red flags people are taught to look for in a cyberattack, such as spelling errors or stilted grammar, are becoming far less reliable.
Cybersecurity experts generally recommend a few baseline habits to reduce exposure to AI-enhanced scams:
- Verify unsolicited investment or financial opportunities through official channels before acting
- Avoid down apps or software recommended through unsolicited messages
- Enable multi-factor authentication on financial and email accounts
- Treat urgency and unusually fast-building trust in online relationships as a warning sign
What Comes Next #
The study does not suggest that AI has made scams unstoppable. Instead, it signals that the fight against a modern cyberattack increasingly depends on how quickly detection tools, public awareness, and platform safeguards can keep pace with offensive AI capabilities.
Researchers involved in the project say further work is needed to understand how these findings apply across different scam formats beyond pig butchering, and how detection systems might be trained specifically to catch AI-generated social engineering attempts before they succeed.
For now, the takeaway for consumers is straightforward. A cyberattack no longer needs a human on the other end of the conversation to be convincing, and the burden of caution is shifting further onto the people receiving the message.