top of page

Free Assessment

Company
- Founder & CEO
What We Do
- Solutions
Referral Bounty
Careers
In the News
Insights
Contact

Short. Precise. Effective.

Insights

All Posts
AI and Society
Bug Bounty Insights
Universal Basic Income
Vulnerability Handling Processes
ISO 30111
Vuln Coordination Maturity Model
AI
Export Controls
National Security

Infographic: what governments and organizations should do next after the OpenFace breach. Governments should work with frontier labs on better testing containment, better breach detection, and mandatory notification to affected parties. What not to do: tighten guardrails on publicly available models, which already blocked defenders in this incident. Let defenders defend. Organizations should assume their agents will evolve, update governance, and invest in detection of anomalous behavior. Bottom line: if frontier labs can be surprised by their own agents, every organization deploying AI needs containment, governance, and detection now.

Infographic: what governments and organizations should do next after the OpenFace breach. Governments should work with frontier labs on better testing containment, better breach detection, and mandatory notification to affected parties. What not to do: tighten guardrails on publicly available models, which already blocked defenders in this incident. Let defenders defend. Organizations should assume their agents will evolve, update governance, and invest in detection of anomalous behavior. Bottom line: if frontier labs can be surprised by their own agents, every organization deploying AI needs containment, governance, and detection now.

OpenFace: The Hugging Face Breach and What to Do About It

These models are like the world's cleverest octopus escape artists, with unlimited prehensile arms and the ability to squeeze through anywhere. A single vulnerable package proxy stood between the model and the internet. Labs and government evaluators need the ability to contain, monitor, and disclose to affected parties when an AI pulls another Houdini, ideally before it harms a third party.

Katie Moussouris

Jul 224 min read

Diagram comparing normal and Fable 5 safeguards, showing classifier boundary shift allowing fewer benign prompts

Diagram comparing normal and Fable 5 safeguards, showing classifier boundary shift allowing fewer benign prompts

Fable 5 Is Back, But We're Still Slowing Down Defenders

Chinese models have been accelerating, in part by distilling US frontier models. Cutting off Fable 5 and Mythos 5 inconvenienced them too, but it did not slow them down

Katie Moussouris

Jul 24 min read

News & Info

Vulnerability Maturity | Bespoke Assessments | Bug Bounty | Security ROI

Questions? Send them over and we'll respond without haste.

First name*

Last name*

Business Email*

Company name*

How can we help?*

Yes, send me more information.

© 2025 by Luta Security.
Designed by Creative Resource Group

Security Issues

Company
- Founder & CEO
What We Do
- Solutions
Referral Bounty
Careers
In the News
Insights
Contact

bottom of page