Microsoft Warns of New AI Chat System Cyber Threat ‘Whisper Leak’

Microsoft Warns of New AI Chat System / Photo File (Reuters)

Microsoft has detected a new cybersecurity vulnerability dubbed “Whisper Leak”, posing a serious risk to the privacy of AI-powered chat systems.

According to the tech giant, this side-channel attack exploits variations in network packet sizes and timing to extract sensitive information, potentially exposing personal conversations, financial data, political discussions, or journalistic communications.

The company clarified that the vulnerability exists in the streaming mode of large language models (LLMs), where chatbots deliver responses in real-time. Differences in packet size and timing can inadvertently reveal clues about the conversation, making it possible for attackers to infer confidential information.

To mitigate the threat, Microsoft has implemented immediate protective measures, including randomizing message content, token batching, and altering network packet structures. These steps make it significantly harder for attackers to reconstruct private chats.

Cybersecurity experts warn that the attack is particularly dangerous for users relying on unsecured networks or platforms under government surveillance. Users are advised to employ VPNs during sensitive conversations and to trust only platforms that have applied the latest security updates.