AI News Feed
Market watch
LG UltraFine Evo 32U990A-S Hits Record-Low $1,099 on Amazon, TechRadar ReportsNewegg Labor Day Deal: 32GB Kingston Fury Beast DDR4-3200 Kit Nearly Half Price at $179.99 with CodeGalaxy Z Fold 8 inner screen corners reported with excessive giveHuawei unveils Mate XT 2 tri-fold with privacy-screen edition as Apple prepares folding iPhoneWall Street Analysts Recommend Nvidia, Uber and Marvell for Long-Term GrowthLenovo, AMD and Huawei Put Local Large-Model AI at Center of New Devices
LG UltraFine Evo 32U990A-S Hits Record-Low $1,099 on Amazon, TechRadar ReportsNewegg Labor Day Deal: 32GB Kingston Fury Beast DDR4-3200 Kit Nearly Half Price at $179.99 with CodeGalaxy Z Fold 8 inner screen corners reported with excessive giveHuawei unveils Mate XT 2 tri-fold with privacy-screen edition as Apple prepares folding iPhoneWall Street Analysts Recommend Nvidia, Uber and Marvell for Long-Term GrowthLenovo, AMD and Huawei Put Local Large-Model AI at Center of New Devices
Cybersecurity

Microsoft Copilot and NASA Ground Software Hit by Serious Security Flaws

Security researchers disclosed vulnerabilities in Microsoft Copilot enabling data theft and in NASA's spacecraft control software allowing unauthorized command execution.

AI News Feed · 2026-08-21 17:58

Security firm Varonis reported that a chain of three vulnerabilities in Microsoft Copilot, together dubbed CoSnitch, could allow attackers to exfiltrate sensitive data. Microsoft tracked the chain as CVE-2026-24301, rated it 8.8 out of 10 in severity, and fixed it with a server-side patch. The researchers said they did not exploit a code bug but used a technique they call 'meta-hacking': by repeatedly asking Copilot about its own guardrails, they mapped its defenses and learned how to bypass them.

In their report, Varonis explained that a maliciously crafted URL, such as 'https://copilot.microsoft.com/?q=&autorun=1', could cause Copilot to execute an attacker-supplied prompt as soon as the link is clicked. When a victim connects Copilot to services like Gmail, Drive, or Calendar, that prompt could instruct the AI to send email addresses, passwords, stored files, and calendar events to the attacker. The third flaw, described as persistent memory poisoning via web summarization, lets a malicious webpage inject instructions into a victim's permanent memory store that survive password changes and session revocations. Varonis noted that this indirect prompt injection stems from the AI's inability to distinguish between instructions and data.

Microsoft was notified of CoSnitch in December 2025 and addressed it in mid-August 2026, according to Varonis. The researchers found no evidence of exploitation in the wild. Varonis warned that the 'meta-hacking' technique could apply to other agentic AI platforms because it exploits a natural language interface rather than a specific code bug.

In a separate disclosure on August 18, researcher Yuval Elbar of Cycode identified a critical vulnerability in NASA's AMMOS Instrument Toolkit GUI (AIT-GUI) versions up to 2.5.1. The browser-based interface runs as a web server on all network interfaces with no authentication, no authorization, and no CSRF protection on changing endpoints. Elbar said an attacker could exploit these failings to execute commands on spacecraft, run server-side scripts, and send files to connected craft. The flaw has been fixed in version 2.5.2.

Cycode advised administrators to upgrade AIT-GUI to 2.5.2, check the console port, and review command history. NASA has not publicly responded to the disclosure. Elbar noted that ground-system software inherits the same web weaknesses as other applications but with a higher cost of failure.