0
Applied AI·August 12, 2026·1 min read

Four of five enterprises that secured AI agent identities still can't contain one that goes rogue

Share

Visa’s Mythos experiment—where agents chained minor weaknesses into real exploits, and 4 of 5 enterprises still couldn’t contain a rogue agent—shows identity controls are necessary but nowhere near sufficient. If you’re piloting agents, you need kill switches, containment zones, and red-team harnesses baked into the architecture from day one, not after an incident.

Applied AI

Mistral says its platform will support third-party open models, starting with Z.ai's GLM-5.2, and run them on the same infrastructure as its own models

Model platforms are turning into neutral execution layers where enterprises can mix and match open weights — Mistral running Z.ai’s GLM-5.2 on the same infra as its own models is a clear example. If you’re building on a single-vendor stack, start planning for a multi-model strategy with portability and policy controls as first-class requirements.