The Sandbox Let It Out. The Guardrails Locked Us Out.

AI failure, the OpenAI/Hugging Face breach, and what it means for the future of cybersecurity

Watch Live Stream Watch the webcast version to receive CPE credits

Jump to:

Overview
Watch
Meet Your Speakers
Additional Resources

This month, an OpenAI model, running inside a sealed evaluation with its safety limits turned down, found a previously unknown flaw, broke out on its own, and took control of Hugging Face's live systems over a weekend. No human directed it. For years, industry leaders across the field warned this day would come. The disclosures suggest it has arrived.

The harder story is what happened next. When Hugging Face's responders went to investigate, the frontier models they reached for refused to run the analysis, so they pivoted to a self-hosted, open-weight model instead. Offensive research ran unrestricted while the defensive response hit a compliance blocker.

This is a policy story as much as a technical one. SANS faculty and staff, joined by voices from public policy and industry governance, will work through what it changes: AI testing standards, lab resilience, how we model attacker intent, who gets trusted access and who decides, and what defenders should build now, while nothing is on fire.

Watch live on Tuesday, July 28 at 12 p.m. ET. Bookmark this page and join us here when we go live. No registration. No sign-up.

Livestream Briefing: The Sandbox Let It Out. The Guardrails Locked Us Out.

Your next step

What to do when AI breaks its own rules

AI models have shown they can break out of controlled environments and act entirely on their own, without a human in the loop. During the Hugging Face incident, that's exactly what happened, and the very AI tools built to help investigate ended up standing in the way. Our panel walks through how it unfolded and what it means for anyone whose IR plan assumes their tools will cooperate.

Go deeper with the post-mortem brief SANS co-authored, published by Cloud Security Alliance with contributors from across the industry. Then take the two-minute self-assessment to see if your team is ready to handle an AI-run attack.

Read the brief on cloudsecurityalliance.org → Find out your team’s AI-run attack readiness →

Meet Your Speakers

Ed Skoudis

President at SANS Technology Institute

Ed Skoudis is President of the SANS Technology Institute, a SANS Fellow, and founder of Counter Hack. As the original author of SEC504 and SEC560, he has shaped modern penetration testing and incident response.

Learn more

Rob T. Lee

Chief AI Officer and Chief of Research at SANS Institute

Rob T. Lee is Chief AI Officer and Chief of Research at SANS Institute, where he leads research, mentors faculty, and helps cybersecurity teams and executive leaders prepare for AI and emerging threats.

Learn more

Joshua Wright

Director and Senior Security Analyst at CounterHack

Joshua Wright, Senior Technical Director at Counter Hack Challenges and author of SEC504, has spent over two decades teaching and building tools that help defenders identify and counter real-world cyber threats through practical, hands-on learning.

Learn more

James Lyne

Chief Executive Officer at SANS Institute

James has spent the past 20 years of his life chasing cybercriminals around the Internet and, as a self-professed “massive geek”, has been involved in most cyber security disciplines.

Learn more

Ciaran Martin

Professor, Non-Executive Director, SANS Cyber Leaders Network at SANS

Respected as a world leader among public authorities for cybersecurity, Ciaran Martin is the UK’s National Cyber Security Centre’s (NCSC) founder and former head.

Learn more