Scientists develop AI tool to weed out gaslighters and narcissists

A team at the University of Huddersfield has created a hybrid AI system that identifies sarcasm, manipulation and abuse patterns – offering new potential for policing and mental health support.

The tool has been developed to help digital forensic teams uncover forms of abuse that are tough to detect online, including patterns of coercive control within large volumes of digital data, such as text messages or chat logs.

Forensic experts say that this kind of “hidden” abuse can be difficult to capture as evidence because it rarely involves a single violent incident.

Evidence of this type of abuse is usually digital, and standard forensic keyword searches often miss its cumulative nature, such as a pattern of gaslighting, isolation, or humiliation.

A team from the UK university’s School of Computing and Engineering sought to address this "data-to-insight gap" by creating a framework they call the Digital Conversation Analysis Pipeline (DCAP).

The framework identifies linguistic indicators of narcissistic abuse, psychological manipulation, and related traits in textual data.

The tool uses what the researchers called a “hybrid artificial intelligence approach,” combining forensic keyword searches with the contextual understanding of a deep learning model.

Narcissist alert

This model enables investigators to detect things like sarcasm and manipulation – even when they are disguised in slang or buzzwords, researchers claim.

A press release issued by the university said that the tool specifically scans for linguistic markers matching the diagnostic criteria of Narcissistic Personality Disorder.

“These are associated with abusive behaviors and include a ‘lack of empathy’ and a ‘strong sense of entitlement’ – vital early-warning signs for investigators tracking patterns of systemic abuse,” it said.

The tool works by acting as a triage system, flagging any abusive patterns and guiding investigators to the most critical evidence.

The tool could help combat online violence against women, especially those who work in public-facing jobs – including human rights defenders, activists, and journalists.

AI tool a “powerful asset” for law enforcement

In a simulated forensic case involving over 8,400 messages, the DCAP tool successfully narrowed the case to 287 key messages containing the most prevalent abusive traits and reduced the investigator's manual review workload by over 90%.

The scientists behind this innovation, PHD researcher Dhruv Patel and senior lecturer Dr. Anju Johnson at the University of Sheffield, claim that the tool has the potential to become a powerful asset for police and forensic investigators.

“The sheer volume of digital data in modern investigations makes it incredibly difficult for human analysts to spot the subtle, long-term linguistic markers of narcissism and psychological abuse. By integrating textual analysis with the evaluation of human emotion via computer vision, we are moving towards a truly multimodal approach."

PHD researcher Dhruv Patel, University of Huddersfield

“This ensures that no nuance is missed, providing an essential framework for both forensic investigations and complex mental health diagnoses,” the researcher added.

Dr. Johnson added that by quantifying the hidden dynamics of personality disorders and abuse, his team has provided prosecutors and the legal system with “a transparent, objective toolkit.”

The Cybernews community is talking about this. Be a part of the conversation.

“This doesn't just clear forensic bottlenecks – it ensures that the voices of the most vulnerable are backed by undeniable, trial-ready data.”

The research has been published in Forensic Science International: Digital Investigation as well as IEEE Access.

Unlock more exclusive Cybernews content on YouTube.