opinion / research essay (Check Point Research, Nov 19, 2026) · published here
Opinion (Check Point Research): a guardrail weaker than the model it protects will eventually be bypassed; the 'gatekeeper asymmetry doom' principle
Practical reading: don't treat a lightweight prompt-injection classifier in front of a capable agent as a security boundary; put the hard limits on what the agent can do (tool scopes, approvals, egress).