What happened
On Sept 30 Bitdefender released AI Guardian as a free public beta for macOS: a background service between an agent and its actions that blocks prompt injection, inspects MCP tools, vets agent skills, and guards credentials/files, returning allowed/flagged/blocked verdicts with auditable logs (prompt analysis runs on-device).
Why it matters
It brings enterprise-style agent runtime guardrails to individual agent users, backed by Bitdefender research showing a 36.5% average tool-poisoning success rate across 20 agents and 1.2M+ exposed AI secrets in 2025 — a notable widening of the agent-security market beyond the enterprise.
Applicability
Individual developers and small teams using Claude Code/OpenClaw should trial AI Guardian for free as an endpoint-level agent runtime guard; enterprises can watch it as a signal for consumer-grade agent security expectations.