Staff Product Security Leader in energy-tech. I secure the software factory - and, lately, the AI we're wiring into it.
I work on "security by design" in the literal sense: threat models that change the architecture before code exists, pipelines that catch what reviews miss, and findings triaged with evidence instead of severity labels. Based in Al Khobar, working across complex energy-tech and digital products.
These repositories are where I think in public. They follow one thread: end-to-end application security, extended to the newest part of the attack surface - the models, the context they consume, and the agents we're starting to trust with real work.
- Securing AI systems. LLM security and context engineering - prompt injection, tool misuse, poisoned context. Securing the "brain" and the "senses" of an agent so it doesn't become a high-speed vulnerability generator.
- Agentic DevSecOps. An SDLC where agents write, review, and ship code inside guardrails: sandboxed CI/CD, MCP servers, LangChain pipelines, autonomous agents with security consultants in the loop.
- The architect's path. Moving from securing one product to designing systems where security is a property, not a phase - and writing down what I learn on the way.
Application-Security End-to-end application security, from secure design by threat modeling to secure deployment. The reference I wish I'd had when I started: what to do at each SDLC stage, and why.
Securing-Context-Engineering Context is the new attack surface. Securing what an AI reads, retrieves, and remembers - because a model with poisoned senses makes confident, wrong, and fast decisions.
LLM-Security Notes and patterns for securing large language models in production: injection, jailbreaks, data leakage, and the controls that actually hold up.
secchallenge An agentic SDLC built end to end: DevSecOps pipeline, LangChain framework, MCP server integration, autonomous agent engineering, CI/CD sandboxing, Spring Boot microservices. The lab where the other repos get tested against reality.
AI-Projects Building AI projects hands-on - you can't secure what you've never built.
Solution-Architect The working notebook of becoming a solution architect: patterns, trade-offs, and decisions with reasons attached.
| Layer | Usually |
|---|---|
| Design | Threat modeling (STRIDE), abuse cases before user stories |
| Code & dependencies | SAST, SCA, and secrets scanning wired into CI - not bolted on |
| Running applications | DAST and IAST in staging; WAF transparency checked before scans |
| Cloud | GCP - GKE, shared-VPC firewalling, Cloud Armor, IAM discipline |
| AI stack | LangChain, MCP, OpenAI-compatible gateways, OpenClaw, OpenBot, Purple Llama, Hugging Face, LiteLLM, vLLM |
| Evidence | Raw requests, log lines, probes - findings ship with proof |
None of this is a religious position. It's the set of controls I've watched fail often enough to know how they fail.
- A finding without evidence is an opinion. Mine ship with the request, the log line, or the probe output attached.
- Shift left, but verify right. Design reviews kill classes of bugs; scanners catch instances; production tells the truth.
- A fix isn't done when the patch lands. It's done when the rescan is green and the runbook exists.
I write on Medium. Recurring subjects:
- Application security that survives contact with real delivery timelines
- Securing AI systems before they're trusted with production access
- What "security by design" looks like when it's a practice, not a slide
- LinkedIn — in/fazil-khan-cyber
- Medium — @nahklizaf
- Happy to talk about: threat modeling workshops, DevSecOps pipelines, and securing agentic systems
Take actions efficiently and work on continuous improvements.