Model Namespace Reuse: An AI Supply-Chain Attack Exploiting Model Name Trust


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:

  • Pro-Tibet groups targeted with ExileRAT in spy campaign

    February 5, 2019

    Researchers have uncovered a new cyberespionage campaign which is targeting pro-Tibetan individuals in order to distribute the ExileRAT Trojan. On Monday, researchers from Cisco Talos said that the new campaign delivers a malicious Microsoft PowerPoint document containing the Remote Access Trojan (RAT) which is capable of stealing system and personal information, terminating or launching processes, surveillance and the ...

  • The APT Name Game: How Grim Threat Actors Get Goofy Monikers

    February 5, 2019

    What’s in a name? When it comes to advanced persistent threat groups, it is often quite a bit. While their monikers’ may seem whimsical – Fancy Bear, Nomadic Octopus, Ocean Lotus and Darkhotel – the reality is these are not arbitrary names. In fact, many are similar to schoolyard nicknames or a type of shorthand – ...

  • Sophisticated new phishing campaign targets the C-suite

    February 5, 2019

    A new phishing campaign to steal login credentials is being launched on businesses – specifically the C-suite. Researchers at GreatHorn first discovered the campaign which targets senior executives by claiming to be from the company’s CEO. The fake email regards the rescheduling of a board meeting. By following the link from this email and users are greeted with a ...

  • Over 485,000 Ubiquiti devices vulnerable to new attack

    February 4, 2019

    Ubiquiti Networks is working on a fix for a newly discovered security issue affecting its devices that attackers have been exploiting since July last year. The issue impacts over 485,000 devices, according to an internet scan conducted by US cyber-security firm Rapid7. Mass-exploitation attacks were first spotted last week by Jim Troutman, co-founder of internet exchange point ...

  • Metro Bank targeted with 2FA-bypassing SS7 attacks

    February 1, 2019

    Metro Bank has reportedly fallen victim to a sophisticated two-factor authentication (2FA) bypass attack after hackers infiltrated a telecoms firm’s text messaging protocol. The Signalling Systems No. 7 (SS7) protocol is used by telecom firms to coordinate how texts and calls are routed around the world. But according to Motherboard, hackers are more actively exploiting SS7, and ...

  • FBI Mapping ‘Joanap Malware’ Victims to Disrupt the North Korean Botnet

    January 31, 2019

    The United States Department of Justice (DoJ) announced Wednesday its effort to “map and further disrupt” a botnet tied to North Korea that has infected numerous Microsoft Windows computers across the globe over the last decade. Dubbed Joanap, the botnet is believed to be part of “Hidden Cobra“—an Advanced Persistent Threat (APT) actors’ group often known as ...