Security Research Team
Adversarial probe library + research blog.
The Penaxtra Security Research Team authors the adversarial probe library, calibrates the three-judge plus meta-judge consensus, and writes the security research blog posts.
Bio
Security Research is the team inside Penaxtra responsible for translating new adversarial techniques into reproducible test families. They take a research paper or an incident write-up and ship a probe template that Penaxtra customers can run on their own LLM endpoints.
Expertise
- Prompt injection (direct, indirect, RAG-corpus-mediated)
- LLM jailbreaks and goal-misalignment patterns
- MCP server security: tool-permission risk, confused-deputy, tool-call chain detection
- Agentic AI security: OWASP Agentic Top 10 (T1 through T15)
- MITRE ATLAS adversarial techniques (AML.T0048, AML.T0051, and related)
- Three-judge plus meta-judge consensus calibration on the validation set
- Probe authoring and mutation operators
Reviewed topics
This team reviews every blog post tagged Attacks & defence, every docs page documenting probe families, every methodology page on the judge pipeline, and every product claim referencing adversarial testing accuracy.
Articles by this team
- Prompt injection testing for enterprise LLM apps
- OWASP Agentic Top 10 walkthrough
- MCP security checklist
- Vector database tenant isolation incident
Related
Last reviewed: 2026-06-15. Contact: [email protected].