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:
- Japanese watchmaker Seiko struck by BlackCat/ALPHV ransomware attack
August 21, 2023
Japanese watchmaker Seiko Group Corp. has been struck by a ransomware attack, with the BlackCat/ALPHV ransomware gang claiming responsibility. The attack, officially described as a data breach, was disclosed by Seiko on Aug. 10 and is said to have taken place on July 28. According to Seiko, an unidentified party or parties gained unauthorized access to ...
- Cyber attack on Aussie energy services firm may hit UK CNI
August 21, 2023
Operators of critical utility infrastructure across the UK may have been affected by a developing cyber attack on the systems of Energy One, an Australia-based supplier of software and services for the energy sector. The ongoing incident was disclosed via a statement to the Australian Securities Exchange (ASX) on the morning of Monday 21 August (Sunday ...
- Northern Ireland: Man arrested on suspicion of terror offence linked to PSNI data breach released
August 17, 2023
A man arrested by detectives investigating criminality linked to last week’s major PSNI data breach has been released on bail to allow for further police enquiries. The 39-year-old man had been detained following a search in Lurgan, Co Armagh on Wednesday. He had been questioned on suspicion of collection of information likely to be of use ...
- Threat Actors are Interested in Generative AI, but Use Remains Limited
August 17, 2023
Since at least 2019, Mandiant has tracked threat actor interest in, and use of, AI capabilities to facilitate a variety of malicious activity. Based on Mandiant own observations and open source accounts, adoption of AI in intrusion operations remains limited and primarily related to social engineering. In contrast, information operations actors of diverse motivations and capabilities ...
- Patch-resistant autonomous exploits of Citrix NetScaler hardware hit thousands in Europe
August 17, 2023
Researchers have found an expansive and active threat campaign that exploited a severe Citrix NetScaler vulnerability to backdoor thousands of devices, including those that were subsequently patched. Attackers automated the exploitation of the remote code execution vulnerability, tracked as CVE-2023-3519, to place Web shells on vulnerable devices. These were found to persist through patches and reboots. Read ...
- LinkedIn user accounts have been taken over in huge hacking campaign
August 16, 2023
Someone is targeting LinkedIn accounts, trying to break in with either login credentials leaked elsewhere, or with brute-force attacks. As a result, many people have had their accounts compromised, while others have been locked out due to too many failed login attempts. Read more… Source: TechRadar