Tom Bonner

Novel Universal Bypass for All Major LLMs HiddenLayer’s latest research uncovers a universal prompt injection bypass impacting GPT-4, Claude, Gemini, and more, exposing major LLM security gaps.

Announcing our latest attack technique, "Policy Puppetry" - a single, transferable prompt blending structured policy & roleplay that bypasses alignment in frontier AI models. Game-changing for red-teaming!

#AI #GenAI #RedTeam #CyberSecurity

hiddenlayer.com/innovation-h...

Silent Sabotage | HiddenLayer Research In this blog, we show how an attacker could compromise the Hugging Face Safetensors conversion space and its associated service bot.

Our researchers discovered that the Hugging Face PyTorch to Safetensors conversion service could easily be compromised by attackers, who could tamper with models and leak the token used to create pull requests from the official bot.

hiddenlayer.com/research/sil...

Some great work by the team, finding 6 CVEs in ClearML and uncovering a complete attack chain that can be exploited to deploy payloads to end-users.

hiddenlayer.com/research/not...