x.comthreadNick Frichette (@Frichette_n) on X
Datadog Security Labs announcement, posted by Nick Frichette, of research on steering the evaluations of a prompt-injection detector called "Jev". The team's method cut its detection rate from about 83% to about 4% in their tests. They also used it to bypass LangChain's experimental tool-risk gating feature that is based on Jev. The tweet gives no method details and links to the full article.
TL;DR Announces Datadog research that dropped a prompt-injection detector's detection rate from ~83% to ~4% and bypassed LangChain's experimental tool-risk gating.
Organisations named here
- Datadog Security Labs
- Nick Frichette
Technologies named here
- Jev
- LangChain experimental Jev-based tool-risk gating feature
- LangChain
Techniques named here
- Prompt Injection
Related in AI Security
blog.talosintelligence.comtoolIntroducing CAIRN: Frontier tracking for AI-integrated malware (Cisco Talos)
github.comadvisoryFlowise MCP Security Bypass Enables RCE
github.comrepoYesWeHack Claude Kit
jfrog.comarticle