New Prompt Injection Attack Vectors Through MCP Sampling


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:

  • Russia’s Trident Ursa (aka Gamaredon APT) Cyber Conflict Operations Unwavering Since Invasion of Ukraine

    December 20, 2022

    Since Unit 42 last blog in early February covering the advanced persistent threat (APT) group Trident Ursa (aka Gamaredon, UAC-0010, Primitive Bear, Shuckworm), Ukraine and its cyber domain has faced ever-increasing threats from Russia. Trident Ursa is a group attributed by the Security Service of Ukraine to Russia’s Federal Security Service. As the conflict has continued ...

  • XLLing in Excel – threat actors using malicious add-ins

    December 20, 2022

    For decades, Microsoft Office applications have served as one of the most significant entry points for malicious code. Malicious actors have continued to utilize Visual Basic for Applications (VBA) macros, despite automatic warnings to users after opening Office documents containing code. In addition to VBA macros, malicious actors, from cybercrime actors to state-sponsored groups, also exploited ...

  • Diving into an Old Exploit Chain and Discovering 3 new SIP-Bypass Vulnerabilities

    December 20, 2022

    More than two years ago, a researcher, A2nkF, published the details of an interesting exploit chain on the Objective-See blog. He demonstrated the exploit chain from root privilege escalation to SIP-Bypass up to arbitrary kernel extension loading. After diving into the second vulnerability of the exploit chain, Trend Micro researchers found that Apple’s patch for this issue is ...

  • Gatekeeper’s Achilles heel: Unearthing a macOS vulnerability

    December 19, 2022

    On July 27, 2022, Microsoft discovered a vulnerability in macOS that can allow attackers to bypass application execution restrictions imposed by Apple’s Gatekeeper security mechanism, designed to ensure only trusted apps run on Mac devices. We developed a proof-of-concept exploit to demonstrate the vulnerability, which we call “Achilles”. Gatekeeper bypasses such as this could be ...

  • Ukraine’s DELTA military system users targeted by info-stealing malware

    December 19, 2022

    A compromised Ukrainian Ministry of Defense email account was found sending phishing emails and instant messages to users of the ‘DELTA’ situational awareness program to infect systems with information-stealing malware. The campaign was highlighted in a report today by CERT-UA (Computer Emergency Response Team of Ukraine), which warned Ukrainian military personnel of the malware attack. DELTA is ...

  • CVE-2022-41040 and CVE-2022-41082 – zero-days in MS Exchange

    December 19, 2022

    At the end of September, GTSC reported an attack on critical infrastructure that took place in August. During the investigation, experts found that two 0-day vulnerabilities in Microsoft Exchange Server were used in the attack. The first one, later identified as CVE-2022-41040, is a server-side request forgery (SSRF) vulnerability that allows an authenticated attacker to ...