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:
- Hack the Real Box: APT41’s New Subgroup Earth Longzhi
November 9, 2022
In early 2022, Trend Micro investigated an incident that compromised a company in Taiwan. The malware used in the incident was a simple but custom Cobalt Strike loader. After further investigation, however, we found incidents targeting multiple regions using a similar Cobalt Strike loader. While analyzing code similarities and tactics, techniques, and procedures (TTPs), we ...
- VMware warns of three critical holes in remote-control tool
November 9, 2022
VMware has revealed a terrible trio of critical-rated flaws in Workspace ONE Assist for Windows – a product used by IT and help desk staff to remotely take over and manage employees’ devices. The flaws are all rated 9.8 out of 10 in CVSS severity. A miscreant able to reach a Workspace ONE Assist deployment, either ...
- Emotet coming in hot
November 8, 2022
Emotet is a ubiquitous and well-known banking trojan that has evolved over the years to become a very successful modular botnet capable of dropping a variety of other threats. Even after a global takedown campaign in early 2021 disrupted the botnet, it reemerged later that year, rebuilding its infrastructure and becoming highly active in a ...
- Shangri-La hotel data breach likely had ‘minimal’ impact at Singapore ministerial summit
November 8, 2022
A recent data breach that hit eight Shangri-La hotels is unlikely to have a large impact on foreign government delegates who attended a high-level defence summit in Singapore, which was held at the hotel. Hackers claiming to have instigated the attack apparently have made contact with the hotel chain. Shangi-La Group said Friday it received an ...
- Malicious extension lets attackers control Google Chrome remotely
November 8, 2022
A new Chrome browser botnet named ‘Cloud9’ has been discovered in the wild using malicious extensions to steal online accounts, log keystrokes, inject ads and malicious JS code, and enlist the victim’s browser in DDoS attacks. The Cloud9 browser botnet is effectively a remote access trojan (RAT) for the Chromium web browser, including Google Chrome and ...
- LockBit affiliate uses Amadey Bot malware to deploy ransomware
November 8, 2022
A LockBit 3.0 ransomware affiliate is using phishing emails that install the Amadey Bot to take control of a device and encrypt devices. According to a new AhnLab report, the threat actor targets companies using phishing emails with lures pretending to be job application offers or copyright infringement notices. The LockBit 3.0 payload used in this attack ...