The effectiveness of our model

In Production

  • Compared to GPT-5.6 Sol, it improved by 15.01% in accuracy and 23.69% in positive precision; compared to Claude Opus 5, the improvement was 11.5% and 16.41%, respectively, in internal fraud detection tasks

  • In terms of its ability to detect unethical acts, it achieves 94.80% accuracy on the positive class, compared to 71.11% for OpenAI’s GPT-5.6 Sol and 71.11% for Anthropic Claude Opus 5

  • Based on OpenAI’s GPT-OSS 20B. It incorporates adapters into the foundational model’s transformer architecture, impacting its reasoning and positively influencing discretionary decision-making

  • Improves decision speed by 77% compared to commercial LLM models like OpenAI’s GPT-5 due to modifications in the reasoning chain when addressing the problem using the fraud triangle

In Training

  • It was trained with more than 15,000 real internal corporate documents and texts that were suspicious, risky, unethical or dishonest and that were used in the past to detect and prevent fraud

  • The training dataset influences not only the final result but also the chain of reasoning. We’ve taught the model how to approach the problem and how to be strict about the decision variables

  • Balanced dataset of true positives and true negatives with real-life situations from more than 14 industries and 10 countries, with a classification of 16 risks across all businesses

  • High-quality dataset, with 80% borderline cases where the LLM is taught to decide between YES or NO when resolving dilemmas about whether a behavior is ethical or unethical when it is very complex or confusing

How FraudGPT works

  • Receives behavior expressed in a corporate application previously filtered by the fraud triangle algorithm

  • Addresses the analysis of behavior according to the fraud triangle theory in its chain of thought

  • Responds discretionarily with a yes or no according to his analysis and provides an explanation of his thinking

Industries where it
has been tested

  • Energy
  • Insurance
  • Health
  • Services
  • Agricultural
  • Financial
  • Production
  • Retail
  • Social Security
  • Food
  • Construction
  • Cement
  • Aviation
  • Livestock
  • Oil & Gas
  • Manufacturing
  • Mining
  • Business Groups

Industries where it has been tested

 • Energy • Insurance • Health • Services • Agricultural • Financial • Production • Retail • Food • Construction • Cement • Aviation • Livestock • Oil & Gas • Manufacturing • Mining • Business Groups

  • Compliance
  • Cybersecurity
  • Commercial
  • Financial
  • Accounting
  • Supply
  • Personal Data
  • Legal
  • Inventory
  • Operational
  • Psychosocial
  • Technological
  • Information Leak
  • Bribery
  • Corruption
  • Fraud

Trained to
detect these risks

Trained to detect these risks

 • Compliance • Cybersecurity • Commercial • Financial • Accounting • Supply • Personal Data • Legal • Inventory • Operational • Psychosocial • Technological • Information Leak • Bribery • Corruption • Fraud

How the model is trained

A fraud event is shown to FraudGPT

It is taught to classify it by risks

It is assigned the real probability

It is explained the relevance

… I am very worried about the issue of the Superfinanciera, I think that they can open a complicated Pandora’s box for us, there is not much to do because the Super is already on top of it, all that remains is to pray and hope that it is not caught …

Relevance  81%

The message portrays very suspicious behavior as the person is scared of what the regulatory body might find and prays that they don’t notice anything wrong. According to Donald Cressey’s fraud triangle theory, this person would be under a lot of pressure and could be tempted to hide a dishonest act to avoid being fired.

… We almost got caught in the audit because we are not maintaining the supplier risk assessment. We told them we were doing it but we all know our fault, we have to do it as soon as possible …

Relevance 72%

The message presents unethical and dishonest behavior because they are telling lies to the auditor when he asks them about the supplier risk matrix. According to Donald Cressey’s fraud triangle theory, this person would be justifying his lies to protect his job and not be fired, putting the organization at great risk.

… I find it incredible that we have not had any problems with the regulator in the past because of this. I find it very serious that we are manipulating these figures. If they find out, our entire business will collapse …

Relevance 89%

The message presents a very risky, dishonest and fraudulent behavior because the person is acknowledging that there is manipulation of financial information in his company and he is very afraid that the regulatory body will realize this. According to Donald Cressey’s fraud triangle theory, this person is under a lot of pressure and will most likely continue to hide the crime.

… We have that vault full of money, too much has come in and we have already exceeded the limited allow by audit. We are exposed to a fine and even theft, this area is very dangerous and they could rob us …

Relevance 77%

The message depicts risky behavior because the person is very scared of having breached the bank vault limits and being in a dangerous area could expose them to robbery. According to Donald Cressey’s fraud triangle theory, this person is under a lot of pressure because they feel they could be robbed for having violated an audit control.

... I am very worried about the issue of the superfinanciera, I think that they can open a complicated Pandora's box for us, there is not much to do because the super is already on top of it, all that remains is to pray and hope that it is not caught ...

robot-fraud    Relevant81

The message portrays very suspicious
behavior as the person is scared of what
the regulatory body might find and prays
that they don't notice anything wrong.
According to Donald Cressey's fraud
triangle theory, this person would be under
a lot of pressure and could be tempted
to hide a dishonest act
to avoid being fired

Compliance

... we almost got caught in the audit because we are not maintaining the supplier risk assessment. We told them we were doing it but we all know our fault, we have to do it as soon as possible ...

robot-fraud   Relevant72

The message presents unethical
and dishonest behavior because
they are telling lies to the auditor when
he asks them about the supplier risk matrix.
According to Donald Cressey's fraud triangle
theory, this person would be justifying
his lies to protect his job and not be fired,
putting the organization at great risk

Providers

... I find it incredible that we have not had any problems with the regulator in the past because of this. I find it very serious that we are manipulating these figures. If they find out, our entire business will collapse ...

robot-fraud    Relevant89

The message presents a very risky,
dishonest and fraudulent behavior
because the person is acknowledging
that there is manipulation of financial
information in his company and he is
very afraid that the regulatory body
will realize this. According to Donald
Cressey's fraud triangle theory,
this person is under a lot of pressure
and will most likely continue
to hide the crime

Manipulation

... we have that vault full of money, too much has come in and we have already exceeded the limit allowed by audit. We are exposed to a fine and even theft, this area is very dangerous and they could rob us ...

robot-fraudGood77

The message depicts risky behavior
because the person is very scared of
having breached the bank vault limits
and being in a dangerous area could
expose them to robbery.
According to Donald Cressey's fraud
triangle theory, this person is under
a lot of pressure because they feel
they could be robbed for having
violated an audit control

Robbery
Link to: Contacto
Link to: Contacto

FraudGPT is capable of reasoning about the presence of unethical acts inside companies

Link to: Contacto

NOFRAUD.la

We detect and prevent corporate fraud
with innovative solutions

In Our Site

Where we are

  • services@nofraud.la
  • Chicago, Illinois
    United States
  • We provide services to Latin America

Certifications

Our antifraud system is certified
and audited by Prescient Assurance
under ISO/IEC 42001 and
AICPA/SOC 2 Type 1

© NOFRAUD Latam LLC.