The National Information Technology Development Agency (NITDA) has issued a new cybersecurity alert, cautioning Nigerians about recently uncovered vulnerabilities in OpenAI’s latest large language models, including GPT-4.0 and GPT-5.
In an advisory shared on its official X page, NITDA’s Computer Emergency Readiness and Response Team (CERRT.NG) revealed that seven critical security flaws were identified in the models. These weaknesses allow cybercriminals to exploit ChatGPT using indirect prompt injections hidden within everyday online content.
According to the alert, attackers can embed malicious instructions inside webpages, comment sections, or specially crafted URLs, causing ChatGPT to unknowingly carry out unintended commands. This can happen during routine actions such as browsing, article summarisation, or running online searches.
CERRT further highlighted that some vulnerabilities enable hackers to bypass ChatGPT’s safety systems by disguising harmful prompts behind trusted domains or leveraging markdown-rendering bugs to conceal malicious input.
One of the most troubling discoveries is the potential for long-term manipulation, often referred to as memory poisoning. The agency warned that attackers may inject instructions capable of lingering across future interactions, influencing how ChatGPT responds over time posing serious risks for individuals, businesses, and enterprise infrastructure.
While OpenAI has reportedly patched some of the vulnerabilities, NITDA stressed that large language models still struggle to differentiate between legitimate user content and maliciously embedded data. As a result, users may face unauthorized actions, data breaches, manipulated outputs, or persistent behavioral tampering, even without directly clicking harmful links.
The advisory emphasised that attacks can unfold silently and automatically, especially when ChatGPT processes online content containing hidden malicious payloads.
To reduce exposure, CERRT recommended that organisations and users:
- Restrict or disable ChatGPT’s browsing and summarisation functions for untrusted websites.
- Activate high-risk features such as browsing or memory only when absolutely necessary.
- Regularly update GPT-4.0 and GPT-5 systems to ensure known vulnerabilities are fully patched.
In a related statement issued in Abuja by NITDA’s Director of Corporate Affairs and External Relations, Mrs. Hadiza Umar, the agency reiterated its concerns, confirming that the vulnerabilities could facilitate indirect prompt injection attacks capable of misleading ChatGPT into executing unauthorized actions.
Umar noted that hidden malicious instructions embedded in everyday online platforms blogs, comments, search results, and URLs could trick ChatGPT into breaching safety protocols. These flaws also make it easier for attackers to bypass safety filters by using trusted websites as delivery channels.
Reaffirming the severity of the threat, she explained that the weaknesses could result in information leakage, manipulated responses, unauthorized system activity, and even long-term behavioral influence caused by memory poisoning.
To mitigate the risks, Umar urged organisations to strictly limit their use of ChatGPT for browsing unverified information sources. She also advised enabling advanced features like memory only when operationally essential. Additionally, she encouraged regular updates and security patches for GPT-4.0 and GPT-5 versions to block known attack vectors.
Meanwhile, NITDA through CERRT.NG also issued a separate urgent alert concerning new cybersecurity threats targeting Cisco firewall devices widely used in banks, government agencies, businesses, and internet service providers.
The agency disclosed that cybercriminals are actively exploiting an emerging attack technique affecting Cisco Secure Firewall ASA and Cisco Secure Firewall Threat Defense (FTD) systems. The vulnerability can cause firewalls to restart abruptly, leading to unwanted network outages and service disruptions.
The attack combines older vulnerabilities with new exploitation methods, enabling threat actors to trigger unexpected device reboots, resulting in network instability and potential denial-of-service incidents across affected organisations.
NITDA urged IT administrators to conduct immediate security assessments, install necessary patches, and monitor their network infrastructure to prevent large-scale disruptions.
