Now available: Inventory Snapshots Learn more about Inventory Snapshots
Obok
Early access

The AI running
in your cloud has risks too.

AI-SPM extends Obok's security coverage to the AI models you deploy and run in production. Detect misconfigurations, excessive permissions and data exposure risks specific to AI workloads.

Same read-only access as the rest of the platform.
AI model inventory · aws:prod + azure:prod 6 models monitored
SageMaker endpoint: fraud-detection-v3
us-east-1 · ml.m5.xlarge · Public endpoint · No auth required
2 risks
Review →
Bedrock agent: customer-support-bot
us-east-1 · Overprivileged IAM role · Access to 14 S3 buckets
3 risks
Review →
Bedrock model: claude-3-haiku (internal)
us-west-2 · Private endpoint · IAM authenticated · No public exposure
No issues
Details →
Azure OpenAI: gpt-4o (customer-facing)
eastus · Managed identity · Content filters enabled
No issues
Details →
Platforms

Where your AI models live, Obok watches.

AI-SPM covers the managed AI services on the clouds you already use. No extra instrumentation, no separate setup.

Amazon Web Services

  • Amazon SageMaker endpoints and pipelines
  • Amazon Bedrock models and agents
  • Amazon Comprehend and Rekognition
  • AWS Lambda functions invoking AI APIs

Microsoft Azure

  • Azure OpenAI Service deployments
  • Azure Machine Learning endpoints
  • Azure AI Studio managed models
  • Azure Cognitive Services

Google Cloud

  • Vertex AI endpoints and model registry
  • Google Gemini API integrations
  • Cloud AI Platform predictions
  • BigQuery ML models
What we detect

AI introduces risks your existing tools were not built for.

Traditional CSPM tools see EC2 instances and S3 buckets. Obok AI-SPM also sees the models, agents and training pipelines running on top of them.

Prompt injection risk

  • Public-facing endpoints with no input validation
  • Agents with access to sensitive tools or data
  • No rate limiting on inference endpoints
  • System prompt exposure via API responses

Training data exposure

  • Training datasets in public S3 buckets
  • Fine-tuning jobs with unencrypted data
  • PII in datasets used for model training
  • Missing data access logging on training stores

Overprivileged models

  • IAM roles with more access than the model needs
  • Agents with write access to production systems
  • API keys embedded in model configurations
  • Cross-account access from AI workloads

Model supply chain

  • Third-party models without provenance verification
  • Unscanned model weights pulled from registries
  • Dependency risks in ML pipelines
  • Model versioning and rollback gaps

Output and logging gaps

  • Inference endpoints with no request logging
  • No monitoring on model outputs in production
  • Sensitive data in model response logs
  • Lack of audit trail for agent actions

Infrastructure misconfigurations

  • AI endpoints exposed on public subnets
  • GPU instances running without encryption
  • Model artifacts stored without versioning
  • No network isolation between AI workloads
Early access

AI-SPM is available for early access customers.

If you're running AI models in production on AWS, Azure or GCP and want visibility into the security posture of those workloads, we want to talk to you. Early access customers work directly with our team and help shape what gets built next.

  • Direct access to the Obok security team
  • Coverage tailored to your AI stack
  • Influence the product roadmap
  • Early access pricing