Palo Alto Unit 42 research uncovered a fundamental flaw in the AI supply chain that allows attackers to gain Remote Code Execution (RCE) and additional capabilities on major platforms like Microsoft’s Azure AI Foundry, Google’s Vertex AI and thousands of open-source projects. We refer to this issue as Model Namespace Reuse.
Hugging Face is a platform that enables AI developers to build, share and deploy models and datasets. On that platform, namespaces are the identifiers of models, which are Git repositories that are stored on the Hugging Face hub. Hugging Face models contain configurations, weights, code and information to enable developers to use the models.
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:
- A dozen US web servers are spreading 10 malware families, Necurs link suspected
April 4, 2019
Researchers have uncovered over a dozen servers, unusually registered in the United States, which are hosting ten different malware families spread through phishing campaigns potentially tied to the Necurs botnet. On Thursday, researchers from Bromium said they have monitored scams connected to this infrastructure during the May 2018 to March 2019 time period. Five families of banking ...
- This new malware is scanning the internet for systems info on valuable targets
April 3, 2019
A new form of malware is scanning the internet for exposed web services and default passwords in what’s thought to be a reconnaissance operation – one which might signal a larger cyberattack is to come. Researchers at AT&T Alien Labs first spotted the malware in March and have named it Xwo after its primary module name. It’s thought that Xwo ...
- Microsoft Edge and Internet Explorer Zero-Days Allow Access to Confidential Session Data
April 2, 2019
On March 30th, security researcher James Lee disclosed information on two zero-day vulnerabilities present in current versions of Microsoft Edge and Internet Explorer. These vulnerabilities make it possible for confidential information to be shared between websites. A flaw in the same-origin policy for these web browsers, called an Origin Validation Error (CWE-346), allows JavaScript embedded in a malicious ...
- Mobile-First Phishing Kit Targets Verizon Customers
April 2, 2019
As people increasingly go mobile-first in their work and personal lives, cybercrime is keeping up: The latest is a phishing kit that specifically targets Verizon Wireless customers in the U.S. According to Jeremy Richards, a researcher at Lookout Security, the kit pushes phishing links to users via email, masquerading as messages from Verizon Customer Support. These ...
- Google Warns of Growing Android Attack Vector: Backdoored SDKs and Pre-Installed Apps
April 1, 2019
Google is reporting an uptick in efforts by bad actors to plant potentially harmful applications (PHAs) on Android devices via pre-installed apps and by bundling them with system updates delivered over the air. The technique is especially troubling, Google said, because PHAs are often malicious and users have no control over what comes pre-installed on their ...
- Critical Rockwell Automation Bug in Drive Component Puts IIoT Plants at Risk
March 29, 2019
A critical Rockwell Automation flaw could be exploited to manipulate an industrial drive’s physical process and or even stop it. A critical denial-of-service (DoS) vulnerability has been found in a Rockwell Automation industrial drive, which is a logic-controlled mechanical component used in industrial systems to manage industrial motors. The vulnerability was identified in Rockwell Automation’s PowerFlex 525 ...