Skip to main content

1. Description

Bud Sentinel is Bud’s proprietary guardrail provider and security layer that brings secure evaluation and runtime trust to the Bud platform. In Bud AI Foundry, Bud Sentinel powers the guardrails workflow with a curated catalog of probes (over 300+), low-latency enforcement (under 10 ms), and a guided experience to select probes, configure rules, and deploy guardrails to projects and endpoints. Bud Sentinel is licensed separately and syncs probes automatically, so newly released detection capabilities appear in the catalog without manual updates. Bud Sentinel focuses on a zero-trust model lifecycle: models are evaluated in isolated sandboxes, scanned for malicious artifacts, and monitored during inference. This lets teams adopt open models while keeping model artifacts and runtime environments continuously protected.

2. USPs (Unique Selling Propositions)

1. Zero‑trust evaluation for model ingestion

Bud Sentinel validates every model in a sandboxed environment, preventing untrusted artifacts from entering production pipelines.

2. Deep artifact scanning beyond antivirus

Model packages are scanned for executable binaries, suspicious payloads, and malicious metadata before they reach shared storage.

3. Runtime trust with continuous monitoring

Bud Sentinel monitors inference workloads for unusual access patterns, privileged operations, and unexpected network activity, with detailed logs for forensic review.

4. 300+ probes with sub‑10 ms latency

The Bud Sentinel probe catalog delivers high‑accuracy detection across PII, safety, bias, and security categories without slowing down applications.

5. Simple, guided guardrail deployment

Bud Admin’s multi‑step workflow makes it easy to choose Bud Sentinel, pick probes, configure rules, and deploy guardrails to projects and endpoints.

6. Automatic probe synchronization

Bud Sentinel probes and rules sync on a scheduled cadence, ensuring detection coverage stays up to date without manual refreshes.

7. Broad model format coverage

Supports common model packaging formats such as H5, Pickle, Safetensors, and TensorFlow SavedModel to keep security consistent across stacks.

8. Performance‑conscious protection

Bud Sentinel is recommended as the default guardrail provider with high accuracy and low latency for production‑grade safety.

3. Features

3.1 Secure model ingestion pipeline

  • Downloads and preprocesses models inside a locked‑down sandbox.
  • Performs deep scanning across weights, scripts, binaries, and metadata.
  • Blocks and quarantines suspicious artifacts before they enter shared storage.

3.2 Protected model registry and object storage

  • Validated assets are promoted to secure object storage after passing checks.
  • Registry gating prevents untrusted models from reaching downstream consumers.

3.3 Runtime trust monitoring

  • Continuous monitoring during inference detects suspicious file reads, privileged access attempts, or unexpected network connections.
  • Centralized logs enable audit trails and forensic review when anomalies appear.

3.4 Bud Sentinel probe catalog in Guardrails

  • Select Bud Sentinel as a provider and browse its probe catalog from the Guardrails workflow.
  • Search and filter probes by category (PII, Bias, All) and tags.
  • Open a “See more” drawer to review probe details before deployment.

3.5 Deployment‑ready guardrail configuration

  • Selected Bud Sentinel probes carry into the deployment flow for projects and endpoints.
  • PII probes trigger additional configuration steps before deployment.

4. How-to Guides

4.1 Access Bud Sentinel from Guardrails

  1. Log in to Bud AI Foundry.
  2. Click on Guard Rails from the side menu.
  3. Select +Add Guardrail to start the provider workflow.

4.2 Select Bud Sentinel as your provider

  1. In the Select Provider step, choose Bud Sentinel from the Bud list.
  2. Click Next to create the workflow and load the Bud Sentinel probe catalog.

4.3 Search and filter Bud Sentinel probes

  1. Use the search bar to find probes by name, tag, or description.
  2. Click PII, Secrets, Safety or All to focus the list.
  3. Select one or more probes to continue.

4.4 Review probe details

  1. Hover over a probe row to reveal actions.
  2. Click See More to open the detailed probe drawer.
  3. Review the probe description, tags, and coverage before selecting.

4.5 Deploy Bud Sentinel guardrails

  1. After selecting probes, continue to the deployment steps.
  2. Choose a deployment type: Guardrail Endpoint or Add to Existing Deployment.
  3. Pick the destination project (and deployment if applicable).
  4. Name the guardrail profile, description, set guard types and severity.
  5. Click Deploy to activate Bud Sentinel guardrails.

5. FAQ

Q1. What problems does Bud Sentinel solve? Bud Sentinel protects the model supply chain by verifying model assets before they reach production and by monitoring runtime behavior for suspicious activity. Q2. How does Bud Sentinel handle untrusted models? All models are ingested in a sandbox and scanned. Suspicious files are blocked and quarantined, while verified assets are promoted to secure storage and registry access. Q3. Is Bud Sentinel included by default? Bud Sentinel is a proprietary service that requires a separate license. The container image is gated and needs valid credentials before it can be pulled. Q4. How often does Bud Sentinel refresh probes and rules? Bud Sentinel probes and rules automatically sync on a scheduled cadence (every 7 days), so new detection capabilities appear without manual action. Q5. What happens if I select PII probes? Bud Sentinel adds a PII rule configuration step where you can enable or disable specific PII detection rules before deploying. Q6. Can I select multiple probes in one guardrail profile? Yes. Bud Sentinel allows multi‑select across probes, and each probe is carried into the deployment workflow. Q7. What model formats does Bud Sentinel support for secure evaluation? Bud Sentinel supports common formats such as H5, Pickle, safetensors, and TensorFlow SavedModel to keep model ingestion secure across ML workflows. Q8. Does Bud Sentinel monitor models after deployment? Yes. It continuously monitors system, network, and cluster‑level signals to detect unexpected access or behavior during inference. Q9. Is Bud Sentinel difficult to operate? No. The workflow is designed to be one‑click from provider selection through deployment, hiding the underlying security complexity.