Home News NITDA Warns of Security Vulnerabilities in New ChatGPT Versions

NITDA Warns of Security Vulnerabilities in New ChatGPT Versions

0
NITDA Warns of Security Vulnerabilities in New ChatGPT Versions

 

 

The National Information Technology Development Agency (NITDA) has cautioned Nigerian internet users and professionals who rely on artificial intelligence platforms, warning that serious security vulnerabilities have been identified in recent versions of ChatGPT.

In a statement, the agency said the flaws, found in GPT-4 and GPT-5 models, could be exploited by attackers to manipulate AI-generated outputs and potentially gain access to sensitive user data.
NITDA identified seven critical weaknesses, noting that malicious instructions can be secretly embedded in seemingly harmless web content such as social media comments, online articles, or shortened links.

According to the agency, these hidden commands may cause AI systems to execute harmful actions during routine tasks like summarising text or browsing the internet.
Other identified risks include the circumvention of safety filters, the concealment of dangerous content through markdown-related bugs, and “memory poisoning,” a technique that gradually alters an AI model’s behaviour over time. NITDA warned that such exploits could result in data leaks or unauthorised actions.

While OpenAI has reportedly addressed some of the vulnerabilities, the agency said large language models still face challenges in detecting highly disguised malicious instructions.

NITDA therefore urged users to exercise caution when using AI tools, verify outputs independently, and remain vigilant against suspicious online content.