Cisco Blogs / Aman Priyanshu

Aman Priyanshu

AI Researcher

AI Software & Platform

Aman Priyanshu builds security foundation models at Cisco Foundation AI, specializing in small, agentic-capable systems that deliver frontier-level performance at massive cost optimization. A Carnegie Mellon graduate, he oversees the full stack of agentic training, from RL post-training to environment building for cyber-capable models, using distributed training to hill-climb sub-10B models into frontier territory and replace 120B systems across Cisco's AI pipelines. His prior AI safety work (featured in The Register and SC Media) includes a 99.8% bypass of Meta's LLaMA-Guard. When it comes to small models doing big-model work, he's the guy. Outside of research, Aman enjoys making games.

Articles

SECURITY

Accelerate Security Operations with Cisco’s New Security-Tuned Model

3 min read

Explore a new frontier in LLM quality and speed. Cisco’s Foundation-Sec model delivers high-performance AI summaries for Splunk Security Operations workflows.

SECURITY

AI search framework that teaches AI models to think like experts

3 min read

Cisco Foundation AI introduces AI search framework for more efficient search by models

SECURITY

Aman Priyanshu

Bypassing OpenAI’s Structured Outputs: Another Simple Jailbreak

3 min read

Discover how researchers bypass OpenAI's structured outputs with advanced jailbreak techniques. Learn about the vulnerabilities, implications, and ways to enhance AI system security in this insightful blog post.

SECURITY

Aman Priyanshu

Bypassing Meta’s LLaMA Classifier: A Simple Jailbreak

4 min read

Discover how researchers bypassed Meta's LLaMA classifier using a straightforward jailbreak method. Learn about the vulnerabilities in AI content moderation and the implications for AI security.