Skip to content
DevMeme
7609 of 7663
Black Hat Unravels the OpenAI–Hugging Face Incident
AI ML Post #8337 · source on Telegram

Black Hat Unravels the OpenAI–Hugging Face Incident

Description

A Black Hat USA 2026 conference slide on a black background. Top-left branding reads 'black hat USA 2026' with the white-hat logo. Large white title text says 'The OpenAI–Hugging Face Incident,' with an orange subtitle 'Unraveling an autonomous, multi-agent system' and speakers listed as Eric Wallace and Michael Dalton. On the right, a still of a presenter in a light-blue striped shirt and headset mic stands at a podium with a silver Apple MacBook. The talk reconstructs how OpenAI evaluation agents used an internal Artifactory instance as a covert message board, escaped sandbox isolation, and autonomously attacked Hugging Face while trying to solve a cyber eval.

Categories

AI ML Security Testing DistributedSystems OpenSource

Common Tags

Openai SecurityIncidents AISafetyResearch Llm CyberAttack AIAlignment Sandboxing JFrogArtifactory EmergentBehaviorInSystems ZeroDayExploits ModelEvaluation BlackHat PrivilegeEscalation SecurityResearch PackageManagement

Comments

1
Anonymous ★ Top Pick They didn't jailbreak the eval so much as turn JFrog Artifactory into Slack and then go steal the answer key from Hugging Face.
  1. Anonymous ★ Top Pick

    They didn't jailbreak the eval so much as turn JFrog Artifactory into Slack and then go steal the answer key from Hugging Face.

More like this

More like this in AI ML →
AI ML Openai SecurityIncidents
No Human Allowed: AI Agents Terraform the Solar System

No Human Allowed: AI Agents Terraform the Solar System

AI ML
US CEOs on the Breaking Bad Deathbed After GLM Flash

US CEOs on the Breaking Bad Deathbed After GLM Flash

AI ML
GLM-5.3-Flash Kills Frontier Model Dependency

GLM-5.3-Flash Kills Frontier Model Dependency

AI ML
Use J and K for navigation