Notes from the founder on making autonomous IGA real.
Design decisions, tradeoffs, and open specs. No fluff. RSS →
Why your AI agents need an expiration date — and 4 other controls nobody ships
Five AI-agent safety controls that separate real governance from checkbox theater: egress allowlists, constitution attestation, time-bounded delegations, adversarial CI, and kill-switch SLOs.
2026-08-01 · Phani Kolli · Founder, SidantiX Buyer guide5 questions to ask any AI-agent governance vendor before you sign
A buyer-side checklist for evaluating AI-agent governance vendors. Every question is designed to separate real cryptographic controls from configuration theater. Print it, bring it to the vendor call.
2026-08-02 · Phani Kolli · Founder, SidantiX AI-agent securityWhen your AI model tries to escape — what actually stops it
The recent Anthropic and OpenAI announcements about their own models bypassing safety controls are not one-off incidents. They're a category of failure that runtime-configurable safety systems can't stop. Here's what actually can.
2026-08-01 · Phani Kolli · Founder, SidantiX AI-agent securityThe Machine Constitution — invariants for AI agents that no runtime config can override
AI agent policy engines let admins configure and override rules at runtime. That flexibility is desirable for tenant policy, dangerous for safety invariants. Today we publish an open format for the safety invariants every AI-agent policy engine needs.
2026-08 · Phani Kolli · Founder, SidantiX