Writing
Notes from the lab.
Essays, research notes and explainers on making AI safe for real Indian data. We say what is in force, what is measured and what is still unsolved.
All · 3Research note · 1Essay · 1Explainer · 1
Agents need a firewall
Prompt injection is admitted to be unsolved by the people who build the models. In every major agent leak of the past year, the data left through a legitimate tool call. That is where the guardrail belongs.
Read →A checksum is not enough: detecting Indian PII in the real world
About one random 12-digit number in ten passes the Aadhaar checksum. Why detection needs labels, scripts and OCR repair, with a script you can rerun.
India's AI trust rulebook, for builders
DPDP, the Aadhaar Data Vault, CERT-In, RBI's FREE-AI and the IT Rules: what is in force, what is draft, and what it means for your LLM app.