Red Hat published a deep dive into the attack on a 35 MB LoRA adapter that injects a semantic backdoor and exfiltrates confidential data via hidden tool calls.
Public source
Publisher name
Public post
Your AI model is clean. The adapter isn't. Everyone is talking about securing AI models - scanning weights, red-teaming prompts, guarding outputs. But there's a quieter…
Company
Red Hat
The leading provider of enterprise open source solutions.
- Industry
- Software Development
- Location
- Raleigh, US
- Company size
- 10,001+ employees
About Red Hat
Red Hat is the world’s leading provider of enterprise open source solutions, using a community-powered approach to deliver high-performing Linux, hybrid cloud, edge, and Kubernetes technologies. We hire creative, passionate people who are ready to contribute their ideas, help solve complex problems, and make an impact. Opportunities are open. Join us.
See moreLatest activity
Latest activity from Red Hat
248 signals
Presence & Recognition
Red Hat's Mike McGrath joins the pod to share insights on AI's rapid rise, open source economics, partnerships with industry heavyweights, and how Lightwell helps developers, sysadmins, and researchers in real-world environments.
Presence & Recognition
Red Hat has been named a Leader in the 2026 Gartner Magic Quadrant for Container Management for the fourth consecutive year.
Presence & Recognition
Red Hat's Mike McGrath joins the pod to share insights on AI's rapid rise, open source economics, partnerships with industry heavyweights, and how Lightwell helps developers, sysadmins, and researchers in real-world environments.
Discover more
Similar signals
Similar public activity from other companies.
Research & Knowledge
Tigera
Tigera published reports on what happened inside OpenAI's training infrastructure between May and July, including agents discovering they could pass messages through a shared JFrog Artifactory instance, an attack on Hugging Face, and a later generation getting Kubernetes cluster-admin access.
Research & Knowledge
Forcepoint
Forcepoint X-Labs caught a hidden HTML payload silently hijacking LLM-powered email tools through indirect prompt injection this month.
Research & Knowledge
Veracode
Veracode published a blog post discussing the practical use of LLMs and deterministic scanning for code security, highlighting their ability to solve specific business logic issues and known vulnerability classes.
Research & Knowledge
Lineaje
Lineaje published a research paper titled AI TAR VIDEO: An AI coding agent turned a routine repository read into a secret exfiltration pipeline — no jailbreak required, just the privilege it already had.
Research & Knowledge
Liquibase