AI Agents Can Request.
Humans Decide.
The CAM Protocol governs sensitive data access and actions requested by AI agents or people. At least M of N named human approvers must approve, no approver may deny, and the configured veto window must pass before CAM sends one signed call to the customer's system, which releases data or executes the action.
For healthcare, AI infrastructure, and financial services teams assessing governance requirements.
How CAM Works Through MCP and API
AI agents and applications connect through https://api.safeloc.co/mcp over Streamable HTTP or https://api.safeloc.co/v1 over REST. At least M of N named human approvers must approve, no approver may deny, and the configured veto window must pass before CAM sends one signed call to the customer's system. That system releases data or executes the action.
Request Created
An AI agent or person requests sensitive data access or an action through MCP or REST
Request Validated
CAM validates request fields, the HTTPS data endpoint, circle, threshold, expiry, and veto window
Circle Responds
Human circle members receive single-use response tokens out of band
Human Approval and Veto
At least M of N named human approvers must approve. Any denial permanently vetoes the request.
Signed Call to Customer System
After approval, no denial, and passage of the configured veto window, CAM sends one signed call to the customer's system, which releases data or executes the action.
Hash-Chained Audit Log
Every state transition is recorded in a SHA-256 hash-chained audit log.
CAM Protocol Architecture
A provisional patent application covering the CAM Protocol was filed in March 2026. CAM governs sensitive data access and actions requested by AI agents or people through named human approval, a permanent veto on any denial, and a signed call to the customer's system after the configured veto window.
How CAM Governs Requests
- AI agents or people can request sensitive data access or actions.
- At least M of N named human approvers must approve; any denial permanently vetoes.
- After the configured veto window passes, CAM sends one signed call to the customer's system.
- The customer's system releases data or executes the approved action. CAM does not execute actions.
- Each decision is recorded in a tamper-evident, hash-chained audit log.
What is the CAM Protocol?
CAM governs sensitive data access and actions requested by AI agents or people. At least M of N named human approvers must approve; any denial permanently vetoes the request.
At least M of N named human approvers must approve, no approver may deny, and the configured veto window must pass before CAM sends one signed call to the customer's system at data_endpoint_url. That system releases data or executes the action. CAM never executes actions or stores governed data, and records each decision in a tamper-evident, SHA-256 hash-chained audit log. MCP is available at https://api.safeloc.co/mcp, and the REST API base is https://api.safeloc.co/v1.
Core Protocol Properties
- Threshold approvals: at least M of N named human approvers
- Denial veto: any denial permanently vetoes the request
- Request expiry: expires_in_seconds from 10 to 3600, default 90
- Veto window: veto_window_seconds from 0 to 60, default 5
- Tamper-evident, SHA-256 hash-chained decision log
- Email is the default approver notification method; approval links go only to approvers
- Configured veto window must pass before release
- MCP over Streamable HTTP for AI agents
- Direct REST API for system-to-system integrations
- Customer assessment of framework fit
One Protocol. Governed Requests for Data and Actions.
CAM began with privacy-first location verification. The API now governs sensitive data access and actions requested by AI agents or people, with named human approvers deciding whether the customer's system proceeds.
AI agents can request sensitive data access or actions; named humans decide whether the customer's system proceeds
Request Created
POST /v1/requests creates the M-of-N access request
Circle Responds
Human members receive single-use response tokens out of band
Human Approval and Veto
At least M of N named human approvers must approve. Any denial permanently vetoes the request.
Signed Call to Customer System
After approvals, no denial, and passage of the configured veto window, CAM sends one signed call to the customer's system, which releases data or executes the action.
An AI agent may request sensitive data access or an action without a human review step.
CAM governs sensitive data access and actions requested by AI agents or people. At least M of N named human approvers must approve, any denial permanently vetoes the request, and the configured veto window must pass before CAM sends one signed call to the customer's system. That system releases data or executes the action. CAM does not execute actions or store governed data, and each decision is recorded in a tamper-evident, hash-chained audit log.
Cross-Industry Use Cases
Same CAM Protocol API. Different high-stakes enterprise contexts.
Quantify Your Governance Gap
Model the financial exposure CAM controls address, based on IBM's per-record breach cost benchmarks.
1. Set Your Inputs
Choose your industry sector, drag the slider to match your employee count, and select your data sensitivity level.
2. Read the Results
The right panel shows estimated breach exposure CAM eliminates, based on IBM's per-record breach cost benchmarks.
3. Compare Architectures
The bar chart compares an illustrative data-retention model. CAM does not store governed data; decisions remain in its tamper-evident audit log.
This model uses IBM's reported per-record breach benchmark as an illustrative reference. CAM reduces retained-data surface because releases are temporary, event-based, and not persistently stored.
Illustrative benchmark model only. Not a guarantee. Industry breach costs vary by implementation, sector, and incident scope.
illustrative benchmark
Records eliminated from breach surface
Benchmark per record used
Ephemeral / Event-Based
CAM does not store governed data. Each decision remains in a tamper-evident, hash-chained audit log.
SafeLoc's approval-based sharing model does not retain location records. CAM's separate audit log records governance decisions without storing the governed data.
API Data Transparency
What the CAM API processes, and what it explicitly never touches.
What CAM Processes
- • Release-request metadata (type, actor, timestamp)
- • Approver group configurations and thresholds
- • Approval and denial events per request
- • Tamper-evident, SHA-256 hash-chained decision log
- • Configured veto-window parameters
- • Agent identity tokens (scoped, not stored persistently)
What CAM Never Touches
- • The governed data payload itself, such as patient or financial records
- • Persistent location history or behavioral profiles
- • AI model weights, training data, or outputs
- • Third-party tracking or advertising data
- • Biometric identifiers
- • Data sold or shared with any third party for profiling
CAM Design Features for Customer Review
Examples of design features customers may assess against their own requirements. CAM does not confer certification or compliance status.
| Principle | Regulations | CAM Mechanism | Design Rationale |
|---|---|---|---|
| Data handling | GDPR, CPRA, HIPAA | Signed call to customer system | CAM does not store the governed data. Customers assess fit with their own data-handling requirements. |
| Request scope | GDPR, CPRA, HIPAA | Request-scoped authorization | Each request is associated with a defined release and its configured approval policy. |
| Human approval | NIST ZTA, HIPAA | At least M of N named human approvers | Any denial permanently vetoes the request. The configured veto window must pass before release. |
| Time-bounded release | SOC 2, ISO 27001, NIST | Configured veto window and signed call | CAM sends one signed call to the customer's system after the required approvals and veto window. That system releases data or executes the action. |
| Human oversight records | EU AI Act, NIST AI RMF | Human approval and denial decisions | CAM records each decision in a tamper-evident, SHA-256 hash-chained audit log. |
| Audit history | HIPAA, SOC 2, PCI DSS | Tamper-evident, SHA-256 hash-chained log | The log records decisions and is designed to make later alteration detectable. |
Review CAM's Governed Data Release Workflow
Priority access for healthcare, financial services, and AI infrastructure teams.