Waqar Javed

Published four connected preprints on agentic AI security evaluation methodology covering reliability gaps in automated refusal/prompt-injection detectors, and a controlled 7,020-trial study finding that agent framework choice explains ~0.06% of security outcome variance versus ~29% for attack type. All open (CC BY 4.0), code open-source. Curious what other builders in this space are seeing.

https://figshare.com/authors/Waq...

I've been building safelabs-eval an open-source framework for red-teaming and evaluating AI agents aligned to the OWASP LLM Top 10.

Before I launch, I'm genuinely curious about the community's current approach to agent safety:

• Are you doing any adversarial testing on your agents before deploying them?

• Which attack vectors concern you most prompt injection, tool misuse, privilege escalation, data exfiltration?

• Are you using any existing tools, or mostly manual testing?

I built this because I couldn't find anything open-source that worked across LangGraph, CrewAI, AutoGen, OpenAI Agents SDK, and Google ADK without forcing you to change your stack.