AI & agentic security
Agentic AI workflows for AI-SAST and pentesting; Arcanum AI Red Team.
Red teamer, bug hunter, pentester and security researcher.
I help companies build secure systems
Scroll down
Pranit is a security engineer with 10+ years in offensive security, spanning penetration testing, red teaming, application and cloud security, IoT security, vulnerability research and attack surface management. He has secured 300+ applications, earned 20+ Hall of Fame placements, and reported 12+ CVEs and vulnerabilities in open-source AI infrastructure, including vLLM, SGLang and Harbor.
Today he leads product security for an AI inference cloud, threat-modeling every new AI/ML service before it ships and building agentic AI workflows for AI-SAST and pentesting. Earlier he ran red and purple team engagements, threat modeling and lead pentests for Fortune 500 companies, and he still hunts bugs independently.
What I bring to an engagement
Agentic AI workflows for AI-SAST and pentesting; Arcanum AI Red Team.
Red and purple teams at Amazon; OSCP, CREST CRT, SANS SEC565.
Design reviews for every new AI/ML service and API; threat modeling at Apple.
300+ applications secured; 12+ CVEs in vLLM, SGLang and Harbor.
Kubernetes hardening for thousands of GPUs; audits across AWS, GCP, Azure and OCI.
Built Reconator, an AI-powered attack surface management platform, in Go.
A timeline of building and breaking
Crusoe.ai
December 2025 - Present
Amazon
2023 - 2025
Apple
2023
HackerOne · Cobalt
2022 - 2023
CRAC Learning
2022 - Present
Varutra Consulting
2017 - 2022
HackerOne · Bugcrowd · Synack
2016 - Present
Papers and tools that push offensive security forward
An LLM-guided recon engine that plans, runs and adapts attack-surface workflows, turning recon into a feedback loop.
A React2Shell (CVE-2025-55182) foothold in a Next.js app, turned into a honeypot that captured the full Mirai kill chain.
The original Reconator paper: modular Go tooling around one data model, for faster and more thorough attack-surface discovery.
Reported to the projects AI platforms run on: inference engines, image registries, and metrics and log stores.
Sharing knowledge with the security community
Smart-contract vulnerability classes, real-world exploits and the defensive patterns that stop them.
Panelist on bug bounty hunting, pentesting and generative AI’s impact on security.
View post
Methodology, tooling and lessons from hundreds of engagements, in HackerOne’s live Q&A.
Read on HackerOne
End-to-end bug bounty methodology: recon, target selection, exploitation and reporting.
Watch on YouTube
Recon-driven testing: attack-surface mapping, automated pipelines and high-signal tool chaining.
Watch on YouTube
Methodology, real-world findings and lessons across web, cloud and emerging AI attack surfaces.
Watch on YouTubeFormal training and continuous learning
University of Maryland, College Park
Network security, exploitation, reverse engineering and forensics.
Pune Institute of Computer Technology
Built offensive security tooling and web applications.
2026 · Arcanum Information Security
2024 · SANS
2024 · CREST · Credential ID 8933988096
2024 · CREST
2021 · Offensive Security · View credential →
2021 · Hack The Box Pro Labs
2020 · Amazon Web Services
2016 · EC-Council
Engagements, research collaborations, or just to say hi