Resources
AI runtime security, explained
Field guides written from the operator docs we run Runline with - the category, the incidents that define it, and how to evaluate controls. No gating, no lead forms.
July 4, 2026 · 6 min read
What is AI runtime security?
AI runtime security is the enforcement layer that evaluates every AI-agent action at execution time, after the model decides, before the side effect. How it differs from prompt guardrails, LLM gateways, and AI control planes.
ReadJuly 4, 2026 · 7 min read
Nine seconds to data loss: what PocketOS teaches about AI agents
In April 2026 a Cursor agent deleted PocketOS's production database and backups in about nine seconds, after ignoring explicit safety rules. The failure analysis, and the architecture that would have stopped it.
ReadJuly 4, 2026 · 6 min read
Prompt guardrails vs runtime enforcement
Rules in .cursorrules and system prompts are advice the model weighs, not controls it obeys. Where prompt guardrails actually help, where they structurally cannot, and what a real enforcement boundary looks like.
ReadJuly 4, 2026 · 5 min read
Mapping the OWASP LLM Top 10 to runtime controls
A practical mapping from OWASP LLM Top 10 risks (prompt injection, insecure output handling, excessive agency) to enforceable runtime controls at the tool-execution boundary, for security reviews and RFPs.
ReadLooking for evidence instead of explainers?
The proof page publishes our CI scenario gauntlet, 43 PocketOS-class attacks replayed against the policy engine on every commit, with results and honestly disclosed limits.
See the scenario gauntlet