Glass Box
A controlled synthetic workplace for advanced AI practice.
AidGPT is developing a visible training environment where learners can practise more capable AI workflows using fictional records. Publication of connected features follows product acceptance testing.
Synthetic assessed practice. No production connection required. Accountable human decisions.
The assessed-practice boundary
Glass Box is a controlled training approach under development.
- Product state
- Pre-acceptance. Connected-feature publication follows product acceptance testing.
- Data boundary
- The course does not require real organisational or personal data.
- Assessed practice uses fictional and synthetic records.
- No production connection is required.
- An accountable human owns the decision.
Transfer beyond the course
Apply the discipline through authorised routes
The course does not require real organisational or personal data. Learners may later apply the disciplines to approved work through their organisation’s authorised tools, policies and authority.
That later application sits outside assessed practice.
Target concept, subject to product acceptance
Target concept in words
- Fictional records would remain inside purpose-built training services.
- MCP would provide a connection pattern.
- An accountable human would review evidence and own the decision.
Connection language
What MCP means here
MCP is a connection protocol and pattern. It is not itself a security boundary.
Conditional technical assurance
What product acceptance must establish
Acceptance criteria only. This is not evidence that connected controls exist today.
MCP provides the connection pattern between an AI workspace and purpose-built training services. It is not itself a security boundary. Product acceptance must establish isolation, authentication, bounded tools, read-only behaviour, evidence and operational testing before connected features are published.