Meta works to fix errors that can leak AI’s claims for users and the content that has been created

Meta fixed a safety error that allowed Meta Ai Chatbot to reach special demands and responses from AI to other users.

Sandeep Hodkasia, founder of Appsecure Appsecure, told, exclusively that Techcrunch paid it to $ 10,000 in Bug Bounty Bonuses to reveal the private sector that he presented on December 26, 2024.

Meta said in a reform on January 24, 2025, Hodcia said, and no evidence was found that the error was harmful.

Tell Hodkasia Techcrunch that he set the error after examining how Amnesty International dead It allows its registered users to edit their artificial intelligence claims to reunify the text and images. He discovered that when a user edits the directed, the Meta’s rear servers appoint the claim and their response created by artificial intelligence with a unique number. By analyzing the network traffic in its browser during the liberation of Amnesty International’s mentor, Hodcasia could change this unique number and dead servers that will re -respond and generated from the artificial intelligence of another person completely.

Mistakes mean that Meta servers were not realized properly to make sure that the user requests the claim and their response to see them. Hodquacia said that the fast numbers created by Meta servers were “easily guess”, which may allow a harmful actor to scout the original users ’claims by quickly changing the fast numbers using automated tools.

“

The news of the error comes at a time when technology giants are scrambling to launch and refine the products of artificial intelligence Many of the risks of security and privacy Associated with its use.

The independent meta ai application, which He appeared for the first time this year To compete with competing applications such as ChatGPT, it was launched on a rock start after some users In public, I shared unintentionally what they thought was private conversations With chatbot.