UMVA has learned that OpenAI has rolled out a game-changing security feature for ChatGPT, dubbed Lockdown Mode, designed to shield users from the growing threat of "prompt injection attacks".
These insidious attacks occur when malicious actors craft deceptive prompts to trick ChatGPT's language model into following harmful instructions or spilling sensitive information. The hidden prompts are often embedded in web pages and data sources, easily overlooked by humans but still processed by the AI.
UMVA can exclusively reveal that Lockdown Mode will restrict several ChatGPT features when enabled, including live web browsing, retrieval of web-based images, Deep Research, and Agent Mode. Instead, users will be served cached content, a more secure alternative.
Sources have confirmed to UMVA that Lockdown Mode is now available to all ChatGPT account types, including Free, Plus, and Pro plans. However, OpenAI cautions that this feature is not a foolproof solution, as malicious instructions can still be embedded in uploaded files or cached content.
The rollout of Lockdown Mode comes as a response to a string of alarming incidents, including the hijacking of AI browsers, smart home devices, and theft of personal information via manipulated images and calendar events. OpenAI's move aims to bolster user protection and trust in ChatGPT.
As Lockdown Mode becomes available to all users, those who don't see the feature in their settings may have to wait a bit longer for it to roll out. For now, users can rest assured that OpenAI is taking proactive steps to safeguard their interactions with ChatGPT.
