Writing
hi, i'm adesh.
i build security for ai, and i use ai to build security.
founder and ceo of raxIT Labs, in sydney.
raxIT Labs builds security for ai.
raxIT AI finds the ways an agent can be tricked.
on the side, raxIT Studio uses ai for security work.
raxIT AI checks agents before they ship and watches them after.
raxIT Studio runs agents that do the security work nobody has headcount for.
the labs release open-source tools as we go.
we do our research in the open.
benchmarks, attack simulations and the threat intel our products run on.
- can a decision model be your guardrail? oct 2026 We ran 97 real prompt-injection emails through 18 decision models and Bedrock.
- ai security threat intelligence sep 2026 The foundational AI security knowledge base behind raxIT.
- nla audit may 2026 The first credible LLM explainability primitive we have seen in three years, wired into a prompt-engineering loop.
- direphish apr 2026 An incident-response simulation engine that clones your organisation as AI agents, drops a threat actor into the room, and runs the attack dozens of times so you get a probability distribution instead of a single story.
- grayzonebench aug 2025 An open-source safety benchmark for the prompts where the right answer is neither a refusal nor full compliance.
also open source. model-scanner, mcp-oauth-sample, securing-ralph-loop
eight years at aws. the last two building its anz ai security practice.
before that, cisco and hcltech.
- co-wrote the reference threat model for ai and agentic apps, now in aws threat composer 2024
- built the anz ai security consulting practice at aws, from zero 2023
- led the rollout of proactive controls in aws control tower 2023
- the guardrail layer that cleared the first bank genai deployments 2023
- led cisco's apac vpn security team 2012
- network security at hcltech and infosys 2009
i write about agent identity, models in production and ai policy.
- ai agent identity: the complete enterprise guide 1mo ago
- ai agent accountability: who's responsible when it acts? 1mo ago
- apple can protect the audio. what happens to asking? 1mo ago
- 10 things to assume before you run gpt-6 astra or claude fable 5.1 in production 1mo ago
- the list finally caught the god agent 2mo ago
building raxIT AI and raxIT Studio.
testing decision models against bedrock guardrails.
the benchmark is at v1.0.0: 19 systems, 9,975 checks.
thinking about trust boundaries when the boundary is the model.
updated october 2026.
on the side, i learn by building things that aren't security.
right now that's myCocoon, a private ai pregnancy companion for iPhone.
it has a core user base.
it reads what your iPhone already records, like sleep, resting heart rate, kicks and mood, and works it out on the phone itself.
cloud ai is off by default. journal text and photos stay on the phone or in your own iCloud.
live on the app store, built for ios 26.
15 years in security. networks first, then cloud, now ai.
adesh gairola. founder and ceo of raxIT Labs, sydney.
cisco ccie security (emeritus), aws solutions architect, aws security specialty, aws ai practitioner, ipv6 forum certified security engineer.
owasp llm top 10 contributor. b.tech, computer science, college of engineering roorkee (2005–2009).