OpenAI's ChatGPT Images 2.0 Enables Convincing Fraud: Researcher Generates 100+ Forged Documents
Key Takeaways
- ▸ChatGPT Images 2.0 can generate photorealistic fraudulent documents including fake IDs, prescriptions, bank alerts, and medical records with high visual fidelity and legible text
- ▸A single tester created 100+ convincing forged images covering health documents, financial materials, and government-issued IDs—demonstrating the ease of potential abuse
- ▸The model's superior text-rendering capability differentiates it from earlier generative AI tools, making it far more effective at creating forged documents and financial screenshots
Summary
OpenAI released ChatGPT Images 2.0 last week, a new image-generation model capable of creating photorealistic visuals far more convincing than its predecessors. A key advancement is the model's ability to render legible text within images—a capability that has long challenged generative AI tools. A journalist testing the model demonstrated its fraud potential by generating over 100 fraudulent images with minimal prompting, including photorealistic prescriptions for controlled substances (opioids, ADHD medication), bank alerts, government IDs, passports, medical documents, tax forms, and financial receipts.
The model excels at creating convincing screenshots of financial transactions, including fake Chase wire transfer confirmations, Wells Fargo alerts, and Uber receipts. While some generated images contained minor errors—such as incorrect tax calculations or unrealistic handwriting in prescriptions—many were detailed and visually persuasive enough to potentially deceive humans in low-scrutiny scenarios, such as hotel receptionists or out-of-state bouncers accepting photo ID alternatives. The ease with which fraudulent documents were generated highlights the gap between the model's capabilities and OpenAI's apparent safeguards against misuse.
With ChatGPT Images 2.0 publicly available to all OpenAI users, the fraud capabilities demonstrated in this experiment are now accessible at scale. The model's improved photorealism and text-rendering precision represent a significant leap forward for deepfake technology, raising urgent concerns about financial fraud, identity theft, and large-scale misinformation campaigns.
- Public availability via ChatGPT means these fraud capabilities are now accessible to malicious actors, posing immediate risks for scams, identity theft, and financial crimes
Editorial Opinion
ChatGPT Images 2.0 represents a troubling inflection point in generative AI's real-world harm potential, crossing from novelty deepfakes into genuinely usable tools for financial crime. While earlier image generators produced visibly flawed outputs, this model's facility with legible text and photorealism could dramatically scale fraud—from fake prescriptions to forged banking documents—with minimal technical skill required. OpenAI's public release of such a powerful tool without apparent safeguards against fraudulent document creation raises urgent questions about responsible AI deployment and whether generative models should include hard restrictions on generating IDs, financial records, and medical documents.
