Agentic applications are programs that leverage AI agents — software designed to autonomously collect data and take actions toward specific objectives — to drive their functionality. As AI agents are becoming more widely adopted in real-world applications, understanding their security implications is critical.
This article investigates ways attackers can target agentic applications, presenting nine concrete attack scenarios that result in outcomes such as information leakage, credential theft, tool exploitation and remote code execution. To assess how widely applicable these risks are, we implemented two functionally identical applications using different open-source agent frameworks — CrewAI and AutoGen — and executed the same attacks on both. Palo Alto findings show that most vulnerabilities and attack vectors are largely framework-agnostic, arising from insecure design patterns, misconfigurations and unsafe tool integrations, rather than flaws in the frameworks themselves.
Read more…
Source: Palo Alto Unit 42
Sign up for our Newsletter
The latest news and insights delivered right to your inbox.
Related:
- Kernel-memory-leaking Intel processor design flaw forces Linux, Windows redesign
January 2, 2018
A fundamental design flaw in Intel’s processor chips has forced a significant redesign of the Linux and Windows kernels to defang the chip-level security bug. Programmers are scrambling to overhaul the open-source Linux kernel’s virtual memory system. Meanwhile, Microsoft is expected to publicly introduce the necessary changes to its Windows operating system in an upcoming Patch ...
- MacOS LPE Exploit Gives Attackers Root Access
January 2, 2018
A researcher that goes by the handle “Siguza” released details of a local privilege escalation attack against macOS that dates back to 2002. A successful attack could give adversaries complete root access to targeted systems. Siguza released details of the attack on Dec. 31 via Twitter, wishing followers a “Happy New Year” and linked to a ...
- Triton Malware Targets Industrial Control Systems in Middle East
December 15, 2017
Researchers found malware called Triton on the industrial control systems of a company located in the Middle East. Attackers planted Triton, also called Trisis, with the intent of carrying out a “high-impact attack” against an unnamed company with the goal of causing physical damage, researchers said. FireEye’s Mandiant threat research team revealed the existence of the malware ...
- 19-Year-Old TLS Vulnerability Weakens Modern Website Crypto
December 13, 2017
A vulnerability called ROBOT, first identified in 1998, has resurfaced. Impacted are leading websites ranging from Facebook to Paypal, which are vulnerable to attackers that could decrypt encrypted data and sign communications using the sites’ own private encryption key. The vulnerability is found in the transport layer security protocol used for Web encryption. A successful attack could ...
- Why bother cracking PCs? Spot o’ malware on PLCs… Done. Industrial control network pwned
December 12, 2017
Security researchers have demonstrated a new technique for hacking air-gapped industrial control system networks, and hope their work will encourage the development of more robust defences for SCADA-based systems. Air-gapped industrial networks are thought to be difficult if not impossible to hack partly because they are isolated from the internet and corporate IT networks. However, in ...
- Process Doppelgänging: New Malware Evasion Technique Works On All Windows Versions
December 7, 2017
A team of security researchers has discovered a new malware evasion technique that could help malware authors defeat most of the modern antivirus solutions and forensic tools. Dubbed Process Doppelgänging, the new fileless code injection technique takes advantage of a built-in Windows function and an undocumented implementation of Windows process loader. Ensilo security researchers Tal Liberman and Eugene Kogan, who ...