Open-source intelligence has changed how organizations detect concerning behavior. Threats posted to social media, "leakage" of an intent to harm, fixation on past attacks, and signals of suicidal thinking often appear online before they surface elsewhere. But detection is only the first step. Someone must decide how serious the concern is and what to do about it. That responsibility may fall to a school threat assessment team, a workplace security officer, a law enforcement analyst, or a leader concerned about a service member. Cogynt Dialog, Cogility's Threat Assessment application, addresses that gap by pairing an AI-guided interview with a transparent, structured model of violence and suicide risk.
The Challenge: Consistent Judgment Under Pressure
Best practice in threat assessment is structured professional judgment: assessors review research-based risk factors, document the evidence, and reach a judgment informed by that review rather than intuition alone. In practice, assessors are often school counselors, human resources and security staff, analysts, or unit leaders - not forensic specialists - working under time pressure with incomplete information.
This can produce inconsistent assessments. Critical questions - such as whether a person has access to a weapon or has made a plan - may go unasked, and two assessors can reach different conclusions without a clear record of why. The stakes are high for both violence and suicide, and their warning signs often overlap.
The Approach: A Guided Dialog Backed by a Transparent Model
Cogynt Dialog is an AI-enabled behavioral threat assessment application that combines structured behavioral models, proprietary decision logic, and large language models to support professional assessment and decision-making. It is built on SOFIT, a knowledge base of behavioral and contextual risk indicators originally developed for insider-threat assessment and designed for use across organizational settings. Cogynt Dialog works in five steps:
- The assessor describes the case in their own words; no expertise in the SOFIT taxonomy is required.
- Cogynt Dialog asks targeted follow-up questions. It uses large language models guided by SOFIT's indicator definitions and coding rules to focus those questions on important gaps, such as the seriousness of a threat or access to means.
- Cogynt Dialog maps the information to SOFIT risk indicators and records the supporting evidence. A transparent probabilistic model combines them into separate violence and suicide risk levels, giving added weight to corroborated warning behaviors; decisive findings such as attack planning can set a minimum risk level.
- Cogynt Dialog produces a plain-language report: the risk levels, the findings that drove them, and the projected effect of possible interventions.
- The assessor reviews and decides. The assessor can correct or add information at any point, updating the analysis before making the final judgment.
Cogynt Dialog gathers and organizes evidence; it does not make the decision.
How Structured Assessment Complements OSINT
OSINT tools are increasingly effective at finding concerning signals; Cogynt Dialog is designed as the next step. Online threats, leakage, fixation on attackers, and statements about wanting to die correspond directly to indicators in its taxonomy. The application places those findings alongside offline context, such as a recent loss, access to weapons, or a history of grievances, and shows how they bear on violence and suicide risk.
Together, detection and structured assessment move organizations from "we found something concerning" to "here is what it means, here is the evidence, and here is what we will do." The record also documents why a concern was judged low risk, which matters when decisions are reviewed later.
Early Results
The first test use case is school threat assessment. In preliminary testing on realistic school case scenarios, the model's risk levels were compared with independent ratings by an experienced school administrator. The violence risk model showed very strong agreement with the expert ratings; the suicide risk model's agreement was more modest, though still meaningful.
Refinement and further testing of Cogynt Dialog continue, and an independent test with an expanded set of cases is planned.
Responsible Use
Assessing violence or suicide risk demands particular care. Cogynt Dialog is designed around four commitments:
- Human judgment is final. The application supports professionals; it does not make or automate decisions about an individual.
- Every result is explainable. Each risk level can be traced to the findings and evidence behind it, so assessors and reviewers can see and challenge how it was reached.
- Prevention, not punishment. The purpose is to identify people who need help and plan support and safety measures, not to label or discipline.
- Privacy by design. The application works with information the organization lawfully holds and must operate within applicable privacy law and policy.
What's Next
The next phase will test Cogynt Dialog with practitioners and evaluate its performance on new cases. Schools are the first use case, not the limit: because the indicators apply across organizational settings, the same approach can support violence and suicide risk assessment and insider risk in private-sector, government, and law enforcement environments.
We welcome organizations interested in piloting or evaluating Cogynt Dialog. To learn more, contact Cogility.
