NITDA Issues Urgent Warning Over New ChatGPT Vulnerabilities Exposing Users to Data-Leakage Attacks

The National Information Technology Development Agency (NITDA) has released an urgent cybersecurity advisory warning Nigerians about newly discovered vulnerabilities in ChatGPT that could expose users to data-leakage and manipulation attacks.

The advisory, issued through the agency’s Computer Emergency Readiness and Response Team (CERRT.NG), follows increasing concerns about the risks of using AI-powered tools to process untrusted online content.

Seven Vulnerabilities Found in GPT-4o and GPT-5 Models

According to NITDA, cybersecurity researchers identified seven vulnerabilities affecting OpenAI’s GPT-4o and GPT-5 models that make them susceptible to indirect prompt injection, a technique that allows attackers to manipulate AI systems using hidden instructions.

The agency revealed that malicious prompts can be embedded inside webpages, comments, images, or crafted URLs. When ChatGPT interacts with these elements through browsing, summarisation, or search, it may unknowingly execute harmful commands.

“By embedding hidden instructions in webpages, comments, or crafted URLs, attackers can cause ChatGPT to execute unintended commands simply through normal browsing, summarization, or search actions,” NITDA stated.

Some vulnerabilities allow attackers to bypass safety filters by hiding malicious code behind trusted domains, while others exploit markdown rendering flaws, enabling hidden prompts to slip through undetected.

More critically, NITDA warned that certain flaws make it possible for attackers to poison ChatGPT’s memory, forcing the system to retain malicious instructions that influence future conversations and outputs.

While OpenAI has patched parts of the issue, the agency noted that large language models still struggle to distinguish legitimate user intent from hidden attacker inputs.

Potential Risks to Nigerians and Businesses

NITDA cautioned that these vulnerabilities could result in a wide range of cybersecurity threats, including:

• Unauthorised actions carried out by the chatbot
• Unintended exposure of sensitive user data
• Manipulated or misleading outputs
• Long-term behavioural changes caused by memory poisoning

CERRT.NG emphasised that users may trigger these attacks without clicking or directly interacting with malicious content—especially when ChatGPT automatically processes webpages or search results containing hidden instructions.

Safety Recommendations for Users and Organisations

To mitigate risks, NITDA advised Nigerians, businesses, and government institutions to implement the following precautions:

• Limit or disable browsing and summarisation of untrusted websites, especially within corporate environments
• Enable browsing or memory selectively, only when strictly necessary
• Regularly update deployed GPT-4o and GPT-5 systems to ensure all known vulnerabilities are patched
• Exercise caution when using ChatGPT for tasks involving sensitive or regulated information

Background: Earlier eSIM Security Alert

This warning comes months after NITDA issued another major cybersecurity alert regarding a critical eSIM vulnerability affecting smartphones, tablets, wearables, and IoT devices worldwide.

The flaw, linked to the GSMA TS 48 Generic Test Profile (version 6.0 and earlier), exposed over 2 billion devices to risks that allowed attackers to:

• Install malicious applets
• Extract cryptographic keys
• Clone eSIM profiles
• Intercept communications
• Execute persistent device-level control

The agency warned that successful exploitation could lead to stealth backdoors and severe privacy breaches.