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:
- AMD Acknowledges Newly Disclosed Flaws In Its Processors — Patches Coming Soon
March 20, 2018
MD has finally acknowledged 13 critical vulnerabilities, and exploitable backdoors in its Ryzen and EPYC processors disclosed earlier this month by Israel-based CTS Labs and promised to roll out firmware patches for millions of affected devices ‘in the coming weeks.’ According to CTS-Labs researchers, critical vulnerabilities (RyzenFall, MasterKey, Fallout, and Chimera) that affect AMD’s Platform Security ...
- Phishing still number one method for cyber-attacks
March 16, 2018
Microsoft has just released its annual cybersecurity report and it says that phishing is still the most popular way for cyber-criminals to attack, giving security experts everywhere headaches. To create the report, Microsoft scanned more than 400 billion emails, 450 billion authentications and 1.2 billion devices. More than half (53 per cent) of all email threats are phishing ...
- Ransomware: Get ready for the next wave of destructive cyberattacks
February 26, 2018
It might look to be out of the limelight compared to 2017, but it would be foolish to write ransomware off yet, as more attacks using the file-encrypting malware are ahead. High profile incidents like WannaCry, NotPetya and Bad Rabbit made ransomeware infamous last year. WannaCry and NotPetya have since both been attributed to be the work of nation-states – the former to North ...
- Hackers are selling legitimate code-signing certificates to evade malware detection
February 22, 2018
Security researchers have found that hackers are using code-signing certificates more to make it easier to bypass security appliances and infect their victims. New research by Recorded Future’s Insikt Group found that hackers and malicious actors are obtaining legitimate certificates from issuing authorities in order to sign malicious code. Read more… Source: ZDNet
- Bot and drone misuse could lead to cybercrime explosion
February 21, 2018
The rapid development of drones and artificial intelligence is a “game-changer” that will present a serious threat to national security if it isn’t addressed. The assessment, made by 26 experts from institutions including Cambridge and Oxford Universities, warns of the potential for malicious use of artificial intelligence (AI) by rogue states, criminals, and terrorists. The panel forecast ...
- Reported Critical Vulnerabilities In Microsoft Software On the Rise
February 15, 2018
The number of reported vulnerabilities in Microsoft software has mounted from 325 in 2013 to 685 last year, a rise of 111 percent, according to new research. Moreover, there has also been a 54 percent increase in critical Microsoft vulnerabilities since 2016, researchers at Avecto said in their report, which is based on data from Microsoft’s Security ...