Severe ChatGPT Bug Exposes Users to Malicious Image-Based Attacks
A newly identified flaw in ChatGPT has opened the door for attackers to inject harmful SVG (Scalable Vector Graphics) and image files into publicly shared conversations—posing a stealthy threat to users through cleverly disguised phishing attempts and malicious content.
Logged under CVE-2025-43714, the vulnerability impacts all versions of ChatGPT up to March 30, 2025.
Security analysts found that ChatGPT mishandles SVG files embedded in chats. Rather than safely displaying the SVG code inside a code snippet, the platform inadvertently renders it as active HTML when users revisit or share conversations via public links. This misstep creates a persistent cross-site scripting (XSS) vulnerability in an otherwise secure AI environment.
“ChatGPT treats SVG content as inline renderable material rather than escaping it for safe display, opening the door to HTML-based code injection on most modern browsers,” explained cybersecurity researcher zer0dac.
This flaw makes it possible for bad actors to embed realistic-looking, weaponized graphics directly into conversations. These visual traps could easily trick users into clicking, compromising security without raising suspicion.
Health Risks and Hidden Code: How ChatGPT’s SVG Flaw Goes Beyond Traditional Cyber Threats
What makes this vulnerability particularly alarming isn’t just its technical severity—but its potential to physically harm users. Experts warn that threat actors could weaponize SVG files with flashing visual patterns capable of triggering seizures in individuals with photosensitive epilepsy, turning a digital bug into a real-world health risk.
At the root of the issue is the unique nature of SVG files. Unlike static image formats like PNG or JPEG, SVGs are written in XML and can legally contain script elements, including HTML and JavaScript. This design flexibility becomes a liability when SVGs are rendered directly within the browser instead of being treated as inert code.
Inline rendering effectively activates whatever markup or scripting lies inside the SVG, allowing it to run inside the user’s browser environment. In this case, ChatGPT’s behavior grants embedded code the ability to execute without warning, exposing users to cross-site scripting (XSS) attacks that originate from a simple shared chat.
“Because SVGs can house executable scripts, they present a unique attack surface,” noted a parallel analysis from another security disclosure. “When mishandled, they let malicious JavaScript run in the context of unsuspecting users, leading to a broad range of exploit scenarios.
ChatGPT Link-Sharing Disabled Amid Security Concerns—Full Patch Still in the Works
In response to the reported security flaw, OpenAI has temporarily suspended its chat link-sharing feature as an immediate containment measure. However, as of now, a complete resolution that addresses the root cause of the vulnerability has not been rolled out.
Cybersecurity professionals are urging users to remain vigilant, particularly when accessing ChatGPT links from unfamiliar sources. While AI-generated content often carries a perception of trustworthiness, this incident serves as a stark reminder that malicious payloads can still be embedded in plain sight.
The nature of the threat goes beyond code execution. According to experts, even SVGs without JavaScript can still be used to manipulate users visually or emotionally—posing serious risks to those with neurological sensitivities or limited technical awareness.
“Manipulation doesn’t require advanced scripts—just the right visuals delivered at the wrong time,” one researcher emphasized. “That alone is enough to cross ethical and security boundaries.”
This incident underscores a broader shift: as AI interfaces become more deeply woven into how people interact with information, they must also be fortified against the same attack vectors that have plagued the web for decades. The lines between chatbots and traditional web platforms are blurring—and the security stakes are rising.
