Twitxy Saved

HiddenLayer

@hiddenlayersec

HiddenLayer helps enterprises safeguard the AI models behind their most important products with a comprehensive security platform
Austin, TX hiddenlayer.com Joined July 2022
24 Following 903 Followers 475 Posts

Latest posts 5

HiddenLayer @hiddenlayersec
@hiddenlayersec has uncovered EchoGram, a technique capable of manipulating the guardrails that protect leading LLMs like GPT-5. This shows the need for diverse, adaptive, & validated security layers to keep pace with rapidly evolving threats. 👉 hiddenlayer.com/innovation-hub…
www.hiddenlayer.com EchoGram: Bypassing AI Guardrails via Token Flip Attacks HiddenLayer unveils EchoGram, a new attack technique that manipulates AI guardrails protecting LLMs like GPT-4, Claude, and Gemini.
HiddenLayer @hiddenlayersec
HiddenLayer researchers have discovered a simple bypass based on our still-functional Policy Puppetry technique for OpenAI's brand-new Jailbreak and Prompt Injection detection guardrails!

Read more 🔗 hiddenlayer.com/innovation-hub…

#AgenticAI #AgenticRisks #AISecurity
www.hiddenlayer.com OpenAI Guardrails Bypass: The "Self-Policing" LLM Vulnerability Explore how the same AI model can behave differently depending on context, creating hidden risks and unpredictable security outcomes.
HiddenLayer @hiddenlayersec
🚨 The first fully AI-powered cyber attack is here.

Anthropic’s report reveals how criminals used Claude Code to run an entire campaign, from data theft to ransom demands.

🔎 Our deep dive: hiddenlayer.com/innovation-hub…

#AISecurity #CyberSecurity #AIThreats
www.hiddenlayer.com The First AI-Powered Cyber Attack: Anthropic Threat Report Cybercriminals are using AI to run entire operations—from reconnaissance to ransom. Explore Anthropic’s findings and key takeaways.