← Back
VulnerabilityCloudflare Blog·3 months ago

Project Glasswing: what Mythos showed us

In recent weeks, we pointed Mythos and other security-focused LLMs at live code across critical parts of our infrastructure. We share what we observed, the models’ strengths and weaknesses, and what the work around them needs to look like before any of it can scale.

Read full article at Cloudflare Blog

Related Articles

VulnerabilitySecurityWeek·2 days ago

Encrypted Prompts Bypass AI Safety Guardrails in Grok and Gemini

Researchers have identified a new attack technique called Cryptographic Context Injection that encrypts malicious instructions to evade safety guardrails in AI models including Grok and Gemini. The method works by concealing harmful prompts until they are decrypted within a trusted execution environment, effectively bypassing existing content filtering mechanisms. This vulnerability highlights a novel attack vector against popular large language models that defenders should monitor.

Project Glasswing: what Mythos showed us | Threat Hunters Journal