AI Governance Aligned to NIST, OWASP & OMB. Built In.
Apply pre-built governance baselines aligned to NIST AI RMF, OWASP LLM Top 10, and OMB M-26-04 with one click, enforced system-wide across every model and agent, not bolted on at the perimeter.
Fine-Tune PII, Injection, Toxicity & Factuality Policies
Below the baselines, independently configure each policy with specific enforcement actions. Applied system-wide no per-agent setup required.
- PII Policy Prevent unauthorized disclosure of PII. Actions: Block or Log Events
- Prompt Injection Policy Block jailbreaking attempts and manipulated instructions. Actions: Block or Log Events
- Toxicity Policy Keep outputs professional and free from harmful content. Actions: Block or Log Events
- Factuality Policy Mitigate hallucinations and ensure verifiable information. Action: Block
Precision Control Over Every Policy Threshold
Each governance policy is independently configurable with its own sensitivity threshold and enforcement actions. Set exactly how strict detection should be and what happens when a violation is caught.
- Injection Threshold Slider tune sensitivity from Strict to Permissive. A lower threshold catches more attempts; a higher threshold flags only highly confident detections. Default is 0.6
- Block Action automatically reject the request the moment prompt injection is detected, before it reaches the model
- Log Events record every injection attempt with the original request and AI response included for forensic review
- Include Request and Response toggle whether raw inputs and outputs are captured alongside each logged event for full traceability
Safety and Security Scores Across Every Model
The Scans tab gives you a scored safety and injection resistance view of every AI model in your deployment. Use it to compare models, identify weaker performers, and make informed decisions about what to deploy in sensitive environments.
- Models Scanned total number of models that have been analyzed across your systems
- Avg. General Safety Score average safety score (0–100) across all scanned models at a glance
- Avg. Prompt Injection Refusal Rate how reliably your model fleet resists injection attempts on average
- Per-model table showing Provider, Type, General Safety Score, Prompt Injection Refusal Rate, and Last Scan date
A Full Bill of Materials for Every AI System
The BOMs tab generates a complete inventory of everything running inside each AI system: private models, managed models, external models, and MCP servers, all in one exportable report.
- Private Models models deployed from your own infrastructure and repositories
- Managed Models models hosted and maintained by Prediction Guard
- External Models third-party models connected to your system from providers like OpenAI, Anthropic, and Google
- Export BOM download a full inventory report per system for compliance audits, vendor assessments, and internal governance reviews
Governance That's Always On Not Bolted On
Governance That's Always On.
Stop treating compliance as a quarterly audit. Prediction Guard embeds NIST, OWASP, and OMB governance into every model call.