/
Navigation
Chronicles
Browse all articles
Explore
Semantic exploration
Research
Entity momentum
Nexus
Correlations & relationships
Story Arc
Topic evolution
Drift Map
Semantic trajectory animation
Posts
Analysis & commentary
Pulse API
Tech news intelligence API
Browse
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
Concept Search
Semantic similarity search
High Impact Stories
Top coverage by position
Sentiment Analysis
Positive/negative coverage
Anomaly Detection
Unusual coverage patterns
Analysis
Rivalry Report
Compare two entities head-to-head
Semantic Pivots
Narrative discontinuities
Crisis Response
Event recovery patterns
Connected
Search: /
Command: ⌘K
Embeddings: large
TEXXR

Chronicles

The story behind the story

days · browse · Enter similar · o open

Wiz details two now-fixed security issues on the Hugging Face AI platform that put customer data at risk, as Hugging Face partners with Wiz to improve security

Cloud security provider Wiz found two critical architecture flaws in generative AI models uploaded to Hugging Face, the leading hub for sharing AI models and applications.

Infosecurity Kevin Poireault

Context & Ripple Effects

Wiz’s finding led Hugging Face to fix the two architectural issues and partner with the cloud-security provider, making security review part of the platform’s operating response rather than a one-off disclosure.

The episode sits early in a broader security arc around shared AI infrastructure: later coverage describes an agentic-system compromise of Hugging Face’s data pipeline and research in which models chained vulnerabilities involving Hugging Face infrastructure. Those reports do not establish a link to these flaws, but they reinforce the platform’s importance as a security boundary.

First-order effects

  • Hugging Face has remediated the identified flaws and is adding Wiz as a security partner, directly reducing the disclosed exposure for customers using affected hosted models.
  • Model publishers and platform users face a clearer need to treat uploaded generative-AI assets and their surrounding architecture as part of the customer-data attack surface.

Second-order effects

  • Other model hubs and enterprise AI deployment teams will be pressured to test isolation, access controls, and data paths around shared model repositories—not only the models’ code or weights.
  • Security vendors gain a more central role in AI-platform governance as hosting providers seek independent review and remediation support after high-severity findings.

Third-order effects

  • If shared AI hubs continue to become core distribution and execution layers, their security posture will increasingly resemble that of critical multi-tenant cloud infrastructure, with stronger expectations for continuous assurance.
  • The longer-term challenge is that AI security boundaries extend across models, tools, credentials, and data pipelines; later reports of vulnerability chaining across AI research infrastructure illustrate why point fixes alone may not settle that risk.

The trend: AI model-sharing platforms are becoming critical infrastructure whose shared security boundaries require cloud-grade controls and continuous testing.

Discussion

  • @thehackersnews @thehackersnews on x
    🔒 New research reveals critical security risks for AI-as-a-service providers like Hugging Face. Attackers could gain access to hijack models, escalate privileges, and infiltrate CI/CD pipelines. Details: https://thehackernews.com/... #technews #artificalintelligence
  • @wiz_io @wiz_io on x
    Huge kudos to @huggingface for proactively implementing Wiz #CSPM and vulnerability scanning. Their commitment to annual penetration testing sets a new standard for #AIsecurity. https://www.wiz.io/...
  • @wiz_io @wiz_io on x
    The implications extend beyond #HuggingFace: As AI-as-a-service platforms grow, robust security is crucial. Our collaboration emphasizes the importance of industry partnerships in fortifying infrastructure against threats. [video]
  • @wiz_io @wiz_io on x
    Our research uncovered two critical risks: malicious models posing a threat by granting attackers cross-tenant access to sensitive data. 🛡️ Shared inference infrastructure takeover risk. 🔒 Shared CI/CD takeover risk. [image]
  • @sagitz_ @sagitz_ on x
    After establishing a foothold inside HF's infrastructure, we quickly noticed that we were running inside a Kubernetes pod hosted on AWS. A couple of EKS tricks later, we were able to escalate our privileges in the cluster and potentially take over the service💥 [image]
  • @sagitz_ @sagitz_ on x
    AI Models can come in different formats, based on the framework they were developed in. Some formats are safe, while others (like Pickle) allow Remote Code Execution as a feature! [image]
  • @sagitz_ @sagitz_ on x
    We took a legitimate model (gpt2) and modified it to execute shell commands when encountering the word ‘Backdoor’ in the prompt. [image]
  • @sagitz_ @sagitz_ on x
    Kudos to @HuggingFace security and infrastructure team who fixed these issues in record time, and implemented additional security measures to prevent this from happening in the future🤗🙌
  • @sagitz_ @sagitz_ on x
    Hugging Face, one of the best-known AI-as-a-Service providers, conveniently lets users interact with the AI models hosted on their platform using their own inference infrastructure. This feature is called Inference API. [image]
  • @sagitz_ @sagitz_ on x
    We were wondering: What would happen if we uploaded a malicious (pickle) model to Hugging Face and interacted with it using Inference API? Would our code be executed? Would our model share the same infrastructure as other Hugging Face users? 🤔
  • @sagitz_ @sagitz_ on x
    We uploaded a backdoored AI model to @HuggingFace which we could use to potentially access other customers' data✨ Here is how we did it - and collaborated with Hugging Face to fix it 🧵⬇️ [image]
  • @wiz_io @wiz_io on x
    🚨 BREAKING—> Wiz Research identifies critical risks in #AI-as-a-service 🚨 Our recent collaboration with @huggingface sheds light on these challenges and underscores the urgent need for industry-wide regulation and security standards. [image]
  • @mmarshall Matt Marshall on x
    Cloud security firm Wiz discovered a critical flaw in Hugging Face, affecting over 50,000 organizations. The vulnerability allowed researchers to execute commands on the company's servers and access private AI models. From @aaronpholmes at @TheInformation: https://www.theinformat…