Red Teaming Generative AI
Why prompts are payloads, and what an adversarial test plan looks like for systems that 'reason' over untrusted text.
cat ./talks
Talks.
I accept a few speaking invitations a year. The sessions that work best are the ones that match my practice: threat modeling, adversary simulation, secure architecture, secure code review, and the current focus on red teaming AI systems and MAESTRO-aligned ML security.
If any of these feel right for your audience, email me with what you have in mind.
Topics.
Threat Modeling AI Agents (STRIDE Is Not Enough)
Why classical frameworks miss EchoLeak-class attacks, and the five-zone methodology that finds them.
MAESTRO for ML Security
CSA's layered framework, applied. What it catches that classical AppSec doesn't, and how to build it into your ML platform without slowing the team.
Building a Security Practice Around LLM Products
How to stand up the controls, runbooks, and culture for organizations shipping generative AI features, without reinventing the wheel.
The Five-Zone Method
Input/retrieval, reasoning, action, state, coordination. A practitioner's map for agentic AI threat modeling that goes beyond enumeration.
Threat Modeling in Smart Manufacturing
A real-world case study on securing Industrial Control Systems (ICS) and IIoT devices at the factory floor boundary without disrupting physical operations.
Where I've spoken.
A few of the recent ones. Happy to talk about any of these in more detail over email.
- Speaker
AI/ML Application Red Teaming
Maharshi Markandeshwar University
Practitioner walk-through of red-teaming methodology for LLM-powered applications and ML systems, with reproducible exercises.
- Speaker
Closed-door briefing on AI/ML security
Industry roundtable (private)
Invited briefing for senior security leaders on threat modeling for agentic AI, the gaps between framework guidance and production deployments, and concrete remediation patterns.
- Interviewee
AI/ML security: interview with a venture capital fund
VC investor briefing
Conversation on the state of AI security tooling, where the market is genuinely under-served, and how technical due diligence should change for AI-native companies.
- Host
Vapra.shiksha: hosted talk series
Vapra.shiksha
Hosted a series of talks on cybersecurity craft and applied AI security for an engineering audience.
Formats.
Keynote
> 30–45 minutes
Conferences, summits, industry events.
Workshop
> ½ or full day
Hands-on, methodology transfer.
Panel
> 45–60 minutes
Industry panels, moderated discussions.
Internal lunch-and-learn
> 60 minutes
For your team, on-site or remote.
For organizers.
Short bio (50 words)
Manish Pandey is a Bengaluru-based Cybersecurity Architect securing AI and agentic systems. He combines 13+ years of hands-on application security craft with an M.Tech in Data Science. He writes on AI red teaming, threat modeling, and how systems that reason fail.
Long bio (150 words)
Manish Pandey is a cybersecurity architect securing AI and agentic systems with the rigor of classical security craft. He has spent 13+ years across the full security stack: application security, secure code review, adversary simulation, and cloud security. He holds an M.Tech in Data Science from BITS Pilani, allowing him to audit the underlying ML math rather than trust the marketing claims. Manish writes and speaks on ML security, agentic AI failure modes, and threat modeling aligned to the CSA MAESTRO framework.
> Need a high-resolution headshot or logos? Email me and I'll send the media kit.
> For organizers: short / long bios above, headshot on request via email.