Meta has patched a major security vulnerability in its AI chatbot platform that could have allowed users to view the private prompts and responses of others. The flaw, discovered by security researcher Sandeep Hodkasia, highlighted a serious privacy concern at a time when big tech firms are racing to push out AI products.
The Discovery
Hodkasia, founder of cybersecurity testing company AppSecure, uncovered the bug while examining how Meta AI handles user-generated prompts. The AI platform, which lets users type in requests for text or image generation, allows people to edit and regenerate their previous inputs.
But during this process, Hodkasia noticed something unusual. When a user edits a prompt, Meta’s back-end servers assign it a unique numerical ID. By intercepting and analyzing network traffic in his browser, Hodkasia realized he could manually change this ID and retrieve prompts and responses belonging to other users entirely.
In essence, the system wasn’t verifying whether the person requesting the data actually had the right to view it. This loophole opened the door for attackers to scrape private conversations at scale by rapidly guessing ID numbers with automated tools.
“The prompt numbers generated by Meta’s servers were easily guessable,” Hodkasia explained. “That meant malicious actors could have exploited it to collect sensitive data.”
The Fix
Hodkasia reported the bug to Meta on December 26, 2024. The company acknowledged the issue, rolled out a fix by January 24, 2025, and rewarded the researcher with a $10,000 bug bounty under its vulnerability disclosure program.
A Meta spokesperson, Ryan Daniels, confirmed to TechCrunch that the company “found no evidence of abuse and rewarded the researcher.”
While Meta insists the bug wasn’t exploited in the wild, the mere existence of such a flaw raises critical concerns about data security in AI platforms that are handling millions of private user prompts daily.
A Pattern of Privacy Concerns
This isn’t the first time Meta AI has faced scrutiny. The chatbot app, launched earlier this year to compete with ChatGPT and other generative AI tools, suffered a rocky rollout. In its early days, some users mistakenly shared private conversations publicly, thinking they were private chats.
The new vulnerability underscores the high stakes of AI security. Unlike traditional apps, AI tools rely on massive data inputs — including user conversations, private prompts, and sensitive context — to function effectively. A breach doesn’t just risk passwords or emails, but potentially deeply personal insights revealed through private queries.
Why This Matters
For Meta, the incident highlights the delicate balance between innovation and security. Big Tech firms are racing to add AI into their products, but flaws like this show that speed can’t come at the cost of safety.
For users, the takeaway is equally important:
-
AI isn’t private by default. Even when conversations feel personal, they may not be securely siloed.
-
Data permanence is real. Once prompts are entered into AI systems, they can remain stored, cached, or exposed through weak security.
-
Vulnerabilities are inevitable. Even the largest companies with billion-dollar infrastructures are not immune.
Security experts argue that companies rolling out AI services need to adopt the same zero-trust mindset used in enterprise security, ensuring that no data is accessible without strict verification.
Industry Context
Meta is not alone in facing such challenges. Earlier reports showed AI-powered apps like Perplexity’s Comet browser and even Microsoft Copilot integrations requesting broad permissions over user data, raising eyebrows in the cybersecurity community.
Signal president Meredith Whittaker recently compared handing your personal data to AI assistants as “putting your brain in a jar,” warning that convenience often comes at the expense of privacy.
With AI platforms becoming more deeply integrated into messaging apps, browsers, and even hardware devices, the risks associated with improper access control or weak backend validation will only grow.
What’s Next for Meta AI
Meta will need to rebuild trust after this incident. While the company’s quick fix and reward to Hodkasia are positive signs, experts believe that transparent security audits and clearer data handling policies are essential.
For now, the $10,000 bug bounty stands as a reminder of how critical independent researchers are in keeping AI platforms accountable.
Final Thoughts
The Meta AI bug may not have been exploited, but it serves as a warning shot: the rush to dominate the AI race can lead to oversights that put users at risk. As AI becomes woven into our daily routines, the need for robust security safeguards has never been greater.
Users should remain cautious, question the permissions they grant, and remember that with AI, convenience often comes with hidden trade-offs in privacy and trust.
