Grant
Specify which external capabilities a given task, user and environment may request. Establish who can approve or revoke a permission and why.
Authority is explicitMASKABLE SUPERINTELLIGENCE AN INDEPENDENT CONCEPT
Possession is
not permission.
Advanced intelligence may be capable of far more than it should be authorized to do. Maskable.si explores a future-facing principle: externally available capabilities should be scoped by explicit policy, independently evaluated and accountable to human authority.
REGISTER / CAPABILITY SURFACE
Select a capability to cycle its illustrated permission state.
SIMULATION ONLY M.SI / 001
MASKED0/ 12
REVIEW REQUIRED0/ 12
EXPOSED0/ 12
↳ Illustrative policy states, not technical enforcement or evidence of safe superintelligence.
This register modifies only local interface state. It does not control an AI model, execute actions or assert that internal model capabilities can be suppressed.
01 — THESIS
Capability is a fact.
Permission is a decision.
THE ASSUMPTION
If a system can, it may.
Capability and authorization collapse into one surface. Every accessible tool or external action becomes a potential route to impact.
THE MASKABLE PROPOSITION
Every external capability has a boundary.
Tool access, operational permissions and delegated actions are evaluated against a defined authority and context—without claiming control over the model's internal knowledge.
WHY THE DISTINCTION MATTERS
More capable AI introduces more consequential decisions about deployment, tool use and autonomy. The strategic question is not whether an advanced model possesses a capacity, but whether a particular operator, task and environment should expose it. A useful control architecture must be measurable and resistant to bypass—not merely described in policy.
02 — MODEL
Grant. Enforce. Verify.
Three proposed design obligations. Each depends on evidence beyond the concept itself.
Enforce
Make unavailable tools and actions unreachable at the execution boundary. Interface restrictions alone are not a substitute for robust enforcement.
Controls must be implementedVerify
Test bypass attempts and failure modes, record evidence and reassess safeguards as systems and capabilities change.
Claims must be tested03 — THE REGISTER
One switch says stop.
A register defines the boundary.
An architectural analogy, not a claim that either mechanism is simple to build or reliable without testing.
| Dimension | Binary stop model | Proposed mask register |
|---|---|---|
| Control surface | System on / off | Permissions for distinct external capabilities |
| Default state | Global allow or stop | Context-specific, least-privilege exposure |
| Granularity | One broad decision | Separate tools, resources and action paths |
| Reversibility | Restart / shutdown | Authorized changes and revocation by policy |
| Evidence | State change is observed | Control efficacy and bypass resistance tested |
04 — SCALE
Maskability levels
A proposed vocabulary for describing how selectively an AI deployment exposes its external capabilities. These levels are not a validated maturity score or an industry standard.
Unbounded
No deliberate capability boundary is expressed in this model.
Coarse
Whole groups of tools are enabled or withheld together.
Granular
Individual tools and action paths receive distinct permissions.
Contextual
Exposure changes with task, environment, role and approval.
Governed
Changes to capability boundaries are logged, reviewed, independently evaluated and subject to revocation. Effectiveness still requires real-world evidence.
Conceptual classification by Maskable.si, independent of any third-party certification or standard-setting body.
05 — EVIDENCE
A serious question.
Testable claims.
The need for capability evaluation and safeguards appears in established frontier-AI research. Maskable's proposed architecture remains a hypothesis for further development.
Citations are independent context, not endorsements of Maskable. No claim is made to solve AI alignment, suppress model-internal capabilities or guarantee control of superintelligence.
06 — THE OPPORTUNITY
Maskable.si is
an idea with room
to become a category.
A distinctive name at the intersection of advanced intelligence, system authority and governance. The concept is open to rigorous development and qualified strategic conversation.
STRATEGIC DIALOGUE
The name. The thesis. The horizon.
For organizations exploring category-defining AI infrastructure, responsible deployment or strategic acquisition.
Strategic inquiries Independent escrow arrangements may be considered by mutual agreement for a domain transaction.