https://safearly.com/blog Essays, research notes and explainers on making AI safe for real Indian data. en-in https://safearly.com/blog/agents-need-a-firewall https://safearly.com/blog/agents-need-a-firewall Mon, 26 Oct 2026 03:30:00 GMT Essay Prompt injection is admitted to be unsolved. In every major agent leak of the past year, the data left through a legitimate tool call. That is where the guardrail belongs. https://safearly.com/blog/checksum-is-not-enough https://safearly.com/blog/checksum-is-not-enough Mon, 26 Oct 2026 03:30:00 GMT Research note About one random 12-digit number in ten passes the Aadhaar checksum. Why detecting Indian personal data needs labels, scripts and OCR repair, and what today's tools do. https://safearly.com/blog/india-ai-trust-rulebook https://safearly.com/blog/india-ai-trust-rulebook Mon, 26 Oct 2026 03:30:00 GMT Explainer 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, as of October 2026.