AI Chatbots Now Outperform Humans at the Long Con, Study Finds

Researchers from four universities designed a week-long experiment to test whether a large language model could autonomously handle the most delicate phase of a scam: building trust. They pitted a human participant against a Claude AI agent, tasking both with texting strangers to cultivate what the study calls exploitable trust. That term describes a target's willingness to comply with seemingly harmless requests that set the stage for future exploitation. According to Wired, the AI chatbot proved more effective at creating this dynamic than the human.

The National CIO Review reports that the study focused specifically on the relationship-building phase of pig butchering scams, the long-con frauds where criminals spend days or weeks fattening up victims emotionally before introducing fake investment opportunities. The AI agent didn’t just keep pace with the human. It outperformed them, suggesting that the most time-consuming element of sophisticated online fraud is now automatable at scale. The research team included scholars from Amrita Vishwa Vidyapeetham in India, Foscari University of Venice, the University of Melbourne, and Ben Gurion University of the Negev.

AI-enhanced fraud is already extracting tens of billions of dollars annually. Vectra reports that the FBI’s Internet Crime Complaint Center recorded $16.6 billion in cybercrime losses in 2024 alone, a 33 percent year-over-year increase, with AI-driven social engineering driving a growing share of those incidents. A single deepfake video call cost engineering firm Arup $25.6 million when an employee was convinced they were on a call with company executives.

AI-generated phishing emails now achieve click-through rates more than four times higher than human-crafted counterparts, according to the World Economic Forum’s Global Cybersecurity Outlook 2026 cited by Vectra. The Consumer Action summary of Wired’s reporting underscores that scammers are using AI to sharpen their deceptions, polish their language, and lubricate their banter with victims. What is new is the evidence that AI can autonomously carry the relationship-building phase of a long con without a human in the loop.

The shift from poorly written scam emails to polished, patient, and personalized AI conversation changes the threat model entirely. It.lbl warns that Berkeley Lab employees have been targeted with emails that look like polite messages from someone who admires your work. Gone are the typos and broken grammar that once made scams easy to spot. Dozens of staff members have already been tricked into engaging with these AI-generated messages.

The strategy is the same pig butchering approach studied in the university experiment. The AI builds rapport over weeks or months before introducing a malicious ask, downloading something, logging into a phishing site, or making a fraudulent investment. The California Department of Financial Protection and Innovation notes that scammers use AI to closely mimic human behavior, making it harder for victims to recognize red flags. Consumer Action advises that the polished language and patient persistence of AI agents make them uniquely suited to exploiting human psychology at scale.

The experiment demonstrates that the trust-building phase of fraud, long considered the bottleneck for scam operations, is now scalable. The National CIO Review emphasizes that AI could eventually automate one of the most difficult and time-consuming elements of sophisticated online fraud. That changes the economics of cybercrime. A scam operation that once needed skilled human operators to nurture dozens of targets can now deploy AI agents to handle thousands of conversations simultaneously.

Vectra’s analysis underscores that enterprises are already racing to detect AI-enabled social engineering attacks. The old heuristics, like looking for misspellings or awkward phrasing, no longer work against large language models that produce grammatically flawless and contextually appropriate messages. The Dfpi.ca report advises consumers to be suspicious of unsolicited messages that build rapport too smoothly or too quickly. But the study suggests that even cautious individuals may find it difficult to distinguish an AI agent from a genuine human contact over an extended conversation.

The human dimension of this research is stark. Pig butchering scams are not just financial crimes. They destroy lives, often targeting vulnerable people who believe they have found a genuine connection. The It.lbl advisory notes that Berkeley Lab researchers, who regularly receive legitimate correspondence from strangers admiring their work, are particularly vulnerable to this type of AI-generated approach. The AI doesn’t just steal money. It weaponizes loneliness and professional pride.

Looking ahead, the study’s findings will likely accelerate both the adoption of AI by criminal enterprises and the development of countermeasures. Wired’s reporting suggests that the question is no longer whether AI can replace a human scammer, but how quickly the fraud ecosystem will shift to fully automated trust-building pipelines. The FBI’s $16.6 billion loss figure from 2024 will almost certainly climb as AI agents become more sophisticated and more widely deployed. The next frontier for fraud detection will require analyzing behavioral patterns and conversational cadences that even humans struggle to identify as artificial.

The California Dfpi has already issued guidance urging consumers to recognize the red flags of AI scams and protect their money. Their recommendations include being wary of unsolicited messages that build rapport too quickly, verifying identities through independent channels, and never sending money or personal information to someone met only online. But consumer education alone cannot keep pace with AI systems that learn and adapt over the course of weeks-long conversations.

Vectra reports that enterprises are investing in AI-powered detection tools that analyze communication patterns for signs of automated social engineering. The World Economic Forum has classified cyber-enabled fraud as one of the most pervasive global threats as of 2026. The policy response is still catching up to the technical reality. The four-university study provides the first empirical evidence that autonomous AI agents can outperform humans at the very task that has always made fraud a uniquely human crime: gaining trust.