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:
- Hacking group ‘8220’ grows cloud botnet to more than 30,000 hosts
July 19, 2022
A cryptomining gang known as 8220 Gang has been exploiting Linux and cloud app vulnerabilities to grow their botnet to more than 30,000 infected hosts. The group is a low-skilled, financially-motivated actor that infects AWS, Azure, GCP, Alitun, and QCloud hosts after targeting publicly available systems running vulnerable versions of Docker, Redis, Confluence, and Apache. Previous attacks ...
- New CloudMensis malware backdoors Macs to steal victims’ data
July 19, 2022
Unknown threat actors are using previously undetected malware to backdoor macOS devices and exfiltrate information in a highly targeted series of attacks. ESET researchers first spotted the new malware in April 2022 and named it CloudMensis because it uses pCloud, Yandex Disk, and Dropbox public cloud storage services for command-and-control (C2) communication. CloudMensis’ capabilities clearly show that ...
- Roaming Mantis hits Android and iOS users in malware, phishing attacks
July 19, 2022
After hitting Germany, Taiwan, South Korea, Japan, the US, and the U.K. the Roaming Mantis operation moved to targeting Android and iOS users in France, likely compromising tens of thousands of devices. Roaming Mantis is believed to be a financially-motivated threat actor that started targeting European users in February. In a recently observed campaign, the threat actor ...
- Botnet malware disguises itself as password cracker for industrial controllers
July 18, 2022
Industrial engineers and operators are being lured into running backdoor malware disguised as tools for recovering access to work systems. These programs offer to crack passwords for specific programmable logic controllers, according to security shop Dragos this month. According to their online ads, the cracking tools can help unlock products from more than a dozen electronics manufacturing ...
- Hackers pose as journalists to breach news media org’s networks
July 16, 2022
Researchers following the activities of advanced persistent (APT) threat groups originating from China, North Korea, Iran, and Turkey say that journalists and media organizations have remained a constant target for state-aligned actors. The adversaries are either masquerading or attacking these targets because they have unique access to non-public information that could help expand a cyberespionage operation. Recent ...
- Meet Mantis – the tiny shrimp that launched 3,000 DDoS attacks
July 15, 2022
The botnet behind the largest-ever HTTPS-based distributed-denial-of-service (DDoS) attack has been named after a tiny shrimp. Cloudflare said it thwarted the 26 million request per second (rps) attack last month, and we’re told the biz has been tracking the botnet ever since. Now, the internet infrastructure company has given the botnet a name — Mantis — ...