AI, with your organisation in control.
AI in DarkRisk is set per tenant. Your organisation's configuration decides which models support your workflows, and our team manages preferences and opt-outs with you.
- 01AI is enabled for this tenantPass
- 02The model is on the tenant's approved listPass
- 03Context is limited to what the task needsApplied
- 04Inputs are excluded from model trainingApplied
- 05The answer goes back for human reviewReview
Your inputs serve your DarkRisk experience.
Relevant context
An AI task may process your prompt, selected documents, questionnaire answers or relevant vendor findings. Our policy is to limit that context to what the task needs.
No general-purpose training
DarkRisk does not use your inputs to train general-purpose AI models. Our policy requires provider terms or hosting safeguards that prevent that use of confidential tenant content.
Tenant-specific choices
Model selection is organisation-scoped. A model listed here is not automatically used for every tenant. Contact us to confirm your tenant's current models and processing arrangements.
How requests reach a model
Direct APIs, AWS services and internal hosting.
DarkRisk integrates with provider APIs directly and uses AWS services for supported deployments. The route depends on the tenant configuration and model availability; not every listed model is available through AWS.
Open-weight models may also be hosted on DarkRisk’s internal infrastructure, subject to their licences, the model’s availability and the tenant’s agreed configuration and restrictions.
For Amazon Bedrock, AWS documents separation from model providers. Retention and abuse monitoring depend on the model, feature and configured controls. A no-training policy does not mean that no data is retained.
The models behind the assistance.
Claude
DeepSeek
Gemini
GLM
GPT
Grok
Hunyuan
Kimi
MiMo
Qwen
One standard for every model
Ask which models your tenant uses- 01No training on your inputs, under the service's terms or our hosting.
- 02Retention that meets your tenant's agreed requirements.
- 03An approved processing location.
- 04Where a developer's own service falls short, an approved hosting route, or no use at all.
Clear rules for responsible use.
- 01Controlled model selection
- AI work uses the tenant's configured models. New model availability does not automatically change a saved configuration, and fallback must stay within the tenant's permitted choices.
- 02Data protection before use
- Our policy requires assessment of confidentiality, training use, retention, security and processing location before confidential content is sent to a service. Being listed here is not approval for every data type.
- 03Human review
- AI can make mistakes. Teams should review generated answers, summaries and recommendations against the evidence before relying on them. AI assistance does not replace accountable vendor-risk decisions.
- 04Clear handling requirements
- Tenant requirements should specify the permitted providers, models, routes and regions. Contact us for the applicable arrangement; this declaration does not promise universal zero retention or in-country processing.
Choose how AI works for your tenant.
Opt out of AI
Request that AI processing is disabled for your tenant. We will identify affected workflows and confirm which features remain available without AI assistance.
Exclude a provider
Request exclusion of a named provider and its model families. Tell us whether the restriction covers all hosting routes, including AWS-hosted versions and models hosted on DarkRisk's internal infrastructure.
Exclude specific models
Keep approved AI assistance while excluding particular model names, versions or IDs. We will confirm suitable alternatives and any effect on feature availability.
Manage your AI preferences
Include your organisation, workspace region and the providers or models you want to exclude. We will verify your authority, agree the change and confirm when it takes effect.
Changes apply from then on. They do not recall completed requests or automatically delete existing records; ask us separately about deletion and retention.