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:
- Utility cyber threats on the rise, but experts say don’t forget basics
July 12, 2023
“There’s an increase in threat actors targeting critical infrastructure,” said Katell Thielemann, a Gartner research analyst focused on risk and security for cyber-physical systems. “And there’s an enhanced sensitivity that threat actors are probing infrastructure.” Since 2021, the U.S. Department of Energy’s annual summary of electric disturbance events shows an uptick in cyber activity. And cyber ...
- Major security flaws in popular Quickblox chat and video framework expose sensitive data of millions
July 12, 2023
Real-time chat and video services available within telemedicine, finance, and smart IoT device applications used by millions of people, rely on the popular QuickBlox framework. QuickBlox supplies mobile and web application developers with a SDK and APIs to deliver not only user management, real-time public and private chat features, for example, but also security features ...
- Hunting for A New Stealthy Universal Rootkit Loader
July 11, 2023
In one of their recent threat hunting investigations, Trend Micro researchers came across an interesting new threat activity cluster that we initially thought was a false positive detection for a Microsoft signed file. However, this turned out to be a novel piece of a signed rootkit that communicates with a large command-and-control (C&C) infrastructure for an ...
- Undocumented driver-based browser hijacker RedDriver targets Chinese speakers and internet cafes
July 11, 2023
Cisco Talos has identified multiple versions of an undocumented malicious driver named “RedDriver,” a driver-based browser hijacker that uses the Windows Filtering Platform (WFP) to intercept browser traffic. RedDriver has been active since at least 2021. RedDriver utilizes HookSignTool to forge its signature timestamp to bypass Windows driver-signing policies. Read more… Source: Talos
- FortiOS/FortiProxy – Proxy mode with deep inspection – Stack-based buffer overflow
July 11, 2023
A stack-based overflow vulnerability in FortiOS & FortiProxy may allow a remote attacker to execute arbitrary code or command via crafted packets reaching proxy policies or firewall policies with proxy mode alongside SSL deep packet inspection. Workaround: Disable deep inspection on proxy policies or firewall policies with proxy mode. Read more… Source: FortiGuard Labs/Fortinet
- Microsoft discloses more than 130 vulnerabilities as part of July’s Patch Tuesday, four exploited in the wild
July 11, 2023
Microsoft released its monthly security update Tuesday, disclosing the most vulnerabilities as part of Patch Tuesday in more than a year. The company released details of more than 130 vulnerabilities, the most in a month since April 2022, 10 of which are considered to be critical. The remaining vulnerabilities are “important.” Read more… Source: Talos