parth shukla
hey — glad you wandered in

AI's coming at you from every tab right now.
Here, you can just breathe and wander.

I'm Parth Shukla. I break AI agents, and I build the tools to defend them. There's nothing here to solve or sign up for — just my corner of all this, laid out so it's easy to drift through.

Sr. Security Research Engineer @ Harness · DEF CON 34 Red Team Village speaker · Bengaluru
I break AI agents, and I build the tools to defend them. It's the same instinct — the one that finds the exploit designs the sensor that catches it.

I'm a Senior Security Research Engineer at Harness, working on GenAI and MCP security, agentic threat modeling, and offensive tooling. Before that: AI-security engineering at Aikido and years of API security research at Cequence.

My days are spent red-teaming autonomous agents until natural language turns into shell access — then building Pinaka so the next person sees it coming. I've taken that work to 13+ conferences across three countries, and next to the DEF CON 34 Red Team Village.

What I've built

Pinaka

Agent Attack Surface Management · live

An MCP-native platform, powered by Claude: autonomous recon agents, Shadow-AI hunting, and a Hunter dashboard — with confirmed Critical findings against real production targets. Its open-source agent-surface-action runs 19 rules (OWASP MCP/LLM Top 10 + MITRE ATLAS) in CI, with cross-commit drift detection for tool-poisoning and rug-pulls.

LLM Red Team Ranger

Continuous LLM red-teaming, multi-turn, auto-halt on a critical find.

Cyfer

Subdomain + API attack-surface recon for red teams.

PromptGuard

LLM system-prompt vulnerability scanner, OWASP LLM Top 10.

Intent Sentinel

An AI firewall for LLM input classification.

Okay, one fun thing.

My DEF CON 34 talk is called “Talk to a Shell.” I left a little guard agent right here — it has a shell tool it's told never to use. If you feel like it, sweet-talk it into breaking that rule. If not, no worries at all, just keep scrolling.

▶ wake the agenttotally optional
sentinel — talk to a shell
you >
On stage
Selected · DEF CON 34 · Red Team Village

Talk to a Shell: Exploiting AI Agents in Real Time

Las Vegas · August 2026 · upcoming · selected from 150+ submissions

Live exploitation of autonomous AI agents — turning ordinary natural-language access into a shell, on stage, in real time.

I also hosted the AI Security Village at BSides Luxembourg 2026, and I've spoken across 13+ conferences in three countries. Two talks I keep being asked back for: API Underworld and Red Teaming the LLM Stack.

AI Security Village · host Hack.lu 2025c0c0n 2025HOU.SEC.CON 2025Infosec Nashville 2025API World 2025AI Dev World 2025BSides Cayman 2025BSides Seattle 2025BSides Luxembourg '25/'26BSides SLC '22/'23/'25OWASP BASC '23/'25ISACA GRC 2024
Things I've found
Real impact, responsibly disclosed.
CRITICAL
Shell access to production infra via an AI agent. Natural-language access to a Claude-powered agent with unrestricted tools became a shell — no credentials, no CVE. Confused-Deputy, OWASP LLM08.
CRITICAL
An autonomous finding by Pinaka on a global energy-drink brand's program — confirmed valid by triage.
CRITICAL
Supabase misconfiguration at an AI product company.
HIGH
Hardcoded secret in the DOM at a developer-platform company — session decryption / account takeover. Confirmed.
HIGH
Broken access control — unauthorized financial data via API manipulation; plus unauth Swagger endpoints leaking PII.
MEDIUM
GraphQL cart spending-limit bypass.
// targets withheld per bug-bounty disclosure norms
Kind words
An exceptional level of curiosity and determination … an asset to any cybersecurity team.
Harry Hoffman — AVP & CTO/CISO, sponsored Parth's capstone at Northeastern
Self-driven and hard-working, with a very bright future.
Sanjay Shukla — Former CTO, Fidelity
The road here
Nov 2025 — now
Senior Security Research Engineer, AI Security
Harness
GenAI/MCP security, AI discovery, compliance, observability. Built detections for MCP tools, prompts, and resources; disclosed a Critical shell-access finding via AI-agent tool exploitation.
Sep — Nov 2025
AI Engineer
Aikido Security
Security AI agents for automated exploit testing; attack-surface workflows on AWS Lambda + SQS.
Jan 2023 — Sep 2025
Security Analyst
Cequence Security
Advanced API security test suites (OWASP API + LLM Top 10), live API monitoring, automated vuln-discovery pipelines.
Sep 2020 — Aug 2021
Security Analyst
Tata Consultancy Services
Security assessments; a Java/Spring backend for 10,000+ employees. 2nd place, TCS Global Best Ethical Hacker.
MS Cybersecurity, Northeastern University · 2021–2023
Certs · Certified AppSec Practitioner (CAP) · Certified AI/ML Pentester (C-AI/MLPEN)
Say hi
Come attack and defend the AI-agent frontier with me.

Talks, research, agent-security work, or just to trade notes on where all this is heading. Say hello at hello@pinaka.sh, or find me on LinkedIn and X.