AI security now treats the model itself as the insider threat.
With frontier models assumed compromised, the industry is rebuilding safety as infrastructure — OS execution containers, agent-identity protocols, and hardened compute substrates — rather than trusting model-level alignment.
The same conclusion keeps arriving from across the workspace's research — 1 topics independently instantiate this theme. Filter the evidence by where it came from:
The OS vendor ships containment as execution containers and Entra-based agent identity so agents can run under an assume-breach posture built into the platform itself.
The CEO-level doctrine rebuilds deployment security around tamper-proof evidence, separation of intelligence from authority, and external kill switches rather than trust in model alignment.
Anthropic's answer to self-replication risk is hardening the device and compute substrate, not model policy.
Poppy gives agents an identity and scoped-permission layer so they stop impersonating users on human-built web pages.