This article examines the security implications of the Model Context Protocol (MCP) sampling feature in the context of a widely used coding copilot application.
MCP is a standard for connecting large language model (LLM) applications to external data sources and tools. We show that, without proper safeguards, malicious MCP servers can exploit the sampling feature for a range of attacks. We demonstrate these risks in practice through three proof-of-concept (PoC) examples conducted within the coding copilot, and discuss strategies for effective prevention.
Read more…
Source: Palo Alto Unit 42
Sign up for the Cyber Security Review Newsletter
The latest cyber security news and insights delivered right to your inbox
Related:
- Italy warns organizations to brace for incoming DDoS attacks
May 30, 2022
Italy’s Computer Security Incident Response Team (CSIRT) has issued an urgent alert to raise awareness about the high risk of cyberattacks against national entities on Monday. The type of cyberattack the Italian organization refers to is DDoS (distributed denial-of-service), which may not be catastrophic but can still cause damage, financial or otherwise, due to service outages ...
- Zero-day vuln in Microsoft Office: ‘Follina’ will work even when macros are disabled
May 30, 2022
Infosec researchers have idenitied a zero-day code execution vulnerability in Microsoft’s ubiquitous Office software. Dubbed “Follina”, the vulnerability has been floating around for a while (cybersecurity researcher Kevin Beaumont traced it back to a report made to Microsoft on April 12) and uses Office functionality to retrieve a HTML file which in turn makes use of ...
- Australian digital driving licenses can be defaced in minutes
May 30, 2022
An Australian digital driver’s license (DDL) implementation that officials claimed is more secure than a physical license has been shown to easily defaced, but authorities insist the credential remains secure. New South Wales, Australia’s most populous state, launched its DDL program in 2019, and as of 2021 officials there said that slightly more than half of ...
- CISA and DoD Release 5G Security Evaluation Process Investigation Study
May 26, 2022
CISA and the Department of Defense (DoD) have released their 5G Security Evaluation Process Investigation Study for federal agencies. The new features, capabilities, and services offered by fifth-generation (5G) cellular network technology can transform mission and business operations; and federal agencies will eventually be applying different 5G usage scenarios: low-, mid-, and high-band spectrum. The study ...
- FBI: Compromised US Academic Credentials Identified Across Various Public and Dark Web Forums
May 26, 2022
The FBI is informing academic partners of identified US college and university credentials advertised for sale on online criminal marketplaces and publically accessible forums. This exposure of sensitive credential and network access information, especially privileged user accounts, could lead to subsequent cyber attacks against individual users or affiliated organizations. Cyber actors continue to conduct attacks against US ...
- New ERMAC 2.0 Android malware steals accounts, wallets from 467 apps
May 26, 2022
The ERMAC Android banking trojan has released version 2.0, increasing the number of applications targeted from 378 to 467, covering a much wider range of apps to steal account credentials and crypto wallets. The goal of the trojan is to send stolen login credentials to threat actors, who then use them to take control of other ...