GUARDBOX | Enterprise AI security
Protect data and usage at the point where AI is used.
GUARDBOX sits between in-house AI applications and external large language models. It is planned to bring traffic audit, sensitive-data masking, data-loss control, access rights and cost management into one entry point.
In development. Planned for release by the end of December 2026.
- Sensitive-data masking
- Data-loss prevention
- Access and cost
- Agent lifecycle
Capabilities
Four areas planned for release
Security gateway
Audit and control traffic between company AI and external models.
Masking and DLP
Handle names, account numbers and similar data before and after the model.
Access and cost
Quotas by department or app, and visibility of token use.
Agent management
Connection, review, runtime and results in one place.
Screen
AI security gateway
Inspect both requests going out and responses coming back, between company AI and external models. Suspicious input, blocks, and the path from an agent to a model are meant to be visible on one screen.
- Inspect prompts and responses both ways
- Block according to rules
- Show the path and status from agent to model
- Mask names, ID numbers and account numbers
Screen
Audit and access records
Keep who did what, and through which agent, in one place. Blocks, flags and allowed actions stay traceable. Quotas by department or application, and token cost, are planned on top of this record.
- One audit log
- Quotas by department or application
- Visibility of token use
- Filter by result and severity
Screen
Data-loss prevention and usage
Watch data going in and coming out, and detect or stop what should not leave. Built for work with strict data rules, such as finance, healthcare and law. Prompt attacks, data loss and unusual permissions are planned on the same status view.
- Detect sensitive data in output
- See data flows, including cross-border
- Trends for prompt attacks and permission issues
- A view of token cost
Screen
Enterprise agent lifecycle
Cover the agents a company uses, from connection approval through security review, runtime monitoring and results. See which applications have how many agents, and which ones need attention.
- Approval and security review before connection
- Risk, rules and runtime in one list
- Agents grouped by application
- Security score and event trend
In development. Planned for release by the end of December 2026.
These screens are from the product in development. You can ask about timing and integration before release.