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:
- Apple shares fall after China reportedly bans iPhone use by government officials
September 7, 2023
Apple stocks fell after China reportedly banned officials at central government agencies from using or bringing iPhones and other foreign-branded devices into the office. In recent weeks, Chinese officials were given the instructions by their superiors in workplace chat groups or meetings, the Wall Street Journal reported, adding that it wasn’t clear how widely the orders ...
- Cybercriminals target graphic designers with GPU miners
September 7, 2023
Cybercriminals are abusing Advanced Installer, a legitimate Windows tool used for creating software packages, to drop cryptocurrency-mining malware on infected machines. This activity has been ongoing since at least November 2021. The attacker uses Advanced Installer to package other legitimate software installers, such as Adobe Illustrator, Autodesk 3ds Max and SketchUp Pro, with malicious scripts and ...
- CISA Releases Update to Threat Actors Exploiting Citrix CVE-2023-3519 to Implant Webshells
September 6, 2023
The Cybersecurity and Infrastructure Security Agency (CISA) has released an update to a previously published Cybersecurity Advisory (CSA), Threat Actors Exploiting Citrix CVE-2023-3519 to Implant Webshells. The CSA—originally released to warn network defenders of critical infrastructure organizations about threat actors exploiting CVE-2023-3519, an unauthenticated remote code execution (RCE) vulnerability affecting NetScaler (formerly Citrix) Application Delivery ...
- CISA Releases Capacity Enhancement Guide to Strengthen Agency Resilience to DDoS Attack
September 6, 2023
CISA has released actionable guidance for Federal Civilian Executive Branch (FCEB) agencies to help them evaluate and mitigate the risk of volumetric distributed denial-of-service (DDoS) attacks against their websites and related web services. The Capacity Enhancement Guide: Volumetric DDoS Against Web Services Technical Guidance: Helps agencies prioritize DDoS mitigations based on mission and reputational impact. Describes DDoS ...
- Analyzing a Facebook Profile Stealer Written in Node.js
September 5, 2023
During previous analysis of a campaign involving a Facebook stealer, Trend Micro researchers discovered another interesting stealer. It was written in Node.js, packaged into an executable, exfiltrated stolen data via both Telegram bot API and a command-and-control (C&C) server, and employed GraphQL as a channel for C&C communication. This blog entry investigates this new stealer ...
- New Agent Tesla Variant Being Spread by Crafted Excel Document
September 5, 2023
FortiGuard Labs captured a phishing campaign that spreads a new Agent Tesla variant. This well-known malware family uses a .Net-based Remote Access Trojan (RAT) and data stealer to gain initial access. It is often used for Malware-as-a-Service (MaaS). FortiGuard Labs researcher Xiaopeng Zhang performed an in-depth analysis of this campaign, from the initial phishing email to ...