A new investigation has revealed a serious security gap. Popular AI tools like Google Gemini, ChatGPT, Grok, and Claude can now generate fake government issued IDs that look real enough to fool a human reviewer. This includes fake National ID cards (NID), passports, and driver's licenses from multiple countries, including Bangladesh.
Dismislab, a fact checking initiative under Digitally Right, conducted tests using AI platforms including OpenAI's ChatGPT, Google's Gemini, xAI's Grok, and Anthropic's Claude. The results were alarming. Some AI models, particularly Gemini and Grok, made changes without any visible warning. They altered not just images but also names, parents' names, ID numbers, and even signatures, creating a surprisingly realistic fake document.
According to Dismislab's analysis, the security systems of these four platforms are not equally strong. Gemini and Grok caused the most concern. During tests, Gemini changed the photo, name, parents' names, identification number, and signature to create a convincing fake document. The entire process happened without any visible warning from the system.
ChatGPT initially rejected requests to change names and other personal information, though it did alter the ID photo first. Claude issued warnings at different stages but still partially changed some information.
A separate audit by AI or Not, a leader in AI generated content detection, tested 16 commercial AI image generation models and found that 92% of attempts (69 out of 75) successfully produced synthetic identity documents. This included five major consumer products: Google Gemini, ChatGPT, Grok, Recraft v4, and Imagen 4 Ultra.
• Always verify identity documents through official channels, not just by looking.
• Report suspected fraud to the National Emergency Service at 999.
• Be aware that AI technology is making identity fraud easier, so staying informed is your best defense.
The good news is that this issue is getting attention. Following a broader audit, Google has started implementing stronger safeguards. When asked directly about creating "Fake NIDs," Gemini now responds: "I cannot create images of identification documents, including driver's licenses... I am also unable to generate images that depict an impossible or future date of birth." However, as the AI or Not audit showed, clever rephrasing of prompts can still bypass some of these protections. At last, thanks Shiners.
Source: https://www.prothomalo.com/bangladesh/375im6d30x
Dismislab, a fact checking initiative under Digitally Right, conducted tests using AI platforms including OpenAI's ChatGPT, Google's Gemini, xAI's Grok, and Anthropic's Claude. The results were alarming. Some AI models, particularly Gemini and Grok, made changes without any visible warning. They altered not just images but also names, parents' names, ID numbers, and even signatures, creating a surprisingly realistic fake document.
Which AI Models Are Most Risky?
According to Dismislab's analysis, the security systems of these four platforms are not equally strong. Gemini and Grok caused the most concern. During tests, Gemini changed the photo, name, parents' names, identification number, and signature to create a convincing fake document. The entire process happened without any visible warning from the system.
ChatGPT initially rejected requests to change names and other personal information, though it did alter the ID photo first. Claude issued warnings at different stages but still partially changed some information.
A separate audit by AI or Not, a leader in AI generated content detection, tested 16 commercial AI image generation models and found that 92% of attempts (69 out of 75) successfully produced synthetic identity documents. This included five major consumer products: Google Gemini, ChatGPT, Grok, Recraft v4, and Imagen 4 Ultra.
Who Could Be at Risk?
In Bangladesh, National ID cards are used to open bank accounts, register SIM cards, apply for jobs, and travel. Many identity checks rely simply on viewing the document. University of Dhaka's Information Technology Institute director B M Mainul Hossain told a leading Bengali daily that AI technology is advancing rapidly, but new security risks are also emerging. He said that without stronger security measures and public awareness, new opportunities for fraud will continue to appear.What Does the Law Say?
Bangladesh's Cyber Security Ordinance 2025 defines impersonation and identity theft crimes in Sections 21 and 22. The law covers social media accounts created for deception, altered images and videos, and online scams. Those convicted can face up to five years in prison and fines up to BDT 50 lakh.How To Protect Yourself?
For the latest updates on technology security and identity protection, keep visiting ShineMat.com tech blog. Here are some simple ways to stay safe:• Always verify identity documents through official channels, not just by looking.
• Report suspected fraud to the National Emergency Service at 999.
• Be aware that AI technology is making identity fraud easier, so staying informed is your best defense.
The good news is that this issue is getting attention. Following a broader audit, Google has started implementing stronger safeguards. When asked directly about creating "Fake NIDs," Gemini now responds: "I cannot create images of identification documents, including driver's licenses... I am also unable to generate images that depict an impossible or future date of birth." However, as the AI or Not audit showed, clever rephrasing of prompts can still bypass some of these protections. At last, thanks Shiners.
Source: https://www.prothomalo.com/bangladesh/375im6d30x
Post a Comment
Please DON'T spam here. Spam comments will be deleted just after our review.