BentoML 1.4

Apply the Python 3 baseline first. These rules cover BentoML 1.4 class-based HTTP Services, dependency services, task endpoints, model stores, and Bento build and deployment workflows, and were checked against 1.4.39. Use a maintained Python runtime supported by the complete model stack; Python 3.10–3.12 overlap this release's published Python classifiers and the baseline. Validate newer interpreters explicitly rather than treating the package's >=3.9 requirement as a compatibility guarantee.

Service and deployment boundaries

• Keep BentoML patched in the build environment, serving image, and dependency services together; do not freeze an old installation indefinitely. BentoML provides security updates only in its latest release.
• Put self-hosted Services behind authentication, authorization, TLS, and admission controls before requests reach BentoML. Set http.host or bentoml serve --host deliberately and restrict direct access to the listener; the default configuration binds 0.0.0.0 and leaves TLS disabled.
• Enable BentoCloud endpoint protection explicitly with access_authorization=True or bentoml deploy --access-authorization true. Give inference clients expiring tokens with Protected Endpoint Access scoped to the required deployments; keep Developer Operations Access credentials out of client applications. Enforce tenant and object authorization in application code beyond the deployment token check.
• Apply authentication and request limits before deserialization through the gateway or Service.add_asgi_middleware(). Cover @bentoml.api,