OSIO is being designed for work that may involve sensitive organisational context and consequential actions. Security and trust therefore depend on clear authority boundaries, controlled execution and honest treatment of uncertainty. The principles below describe design goals; they are not a certification or a guarantee that risk can be eliminated.
Least authority
AI should receive only the authority needed for the bounded work it has been assigned. A broad technical capability should not automatically become permission to use it.
Fresh authority
Consequential actions should not rely blindly on stale assumptions about authority. The permission relevant to an action should still apply when that action is taken.
Credential containment
Secrets and credentials belong behind governed execution boundaries. They should not become ordinary model context or organisational memory.
Untrusted external input
Emails, documents, webpages and provider responses are data to assess, not authority to rewrite OSIO’s rules or grant new permissions.
Governed external effects
Changes in external systems should happen through explicit, bounded execution paths. The fact that an action is technically possible does not establish that it is authorised.
Human judgement
Work should stop and involve a person when genuine judgement or approval is required. The system should not hide uncertainty behind an automatic action.
Evidence over assumption
An execution attempt is not automatically a confirmed outcome. Evidence should support statements about what actually happened.
Replay safety
If an action may already have happened but confirmation is unavailable, the system should avoid blindly repeating a consequential effect. The uncertainty needs to be resolved or handled deliberately.
Isolation
Work and organisational contexts should remain appropriately separated so that information or authority from one context is not casually applied to another.
Fail-closed behaviour
When necessary authority, information or safety conditions are absent, a consequential action should not silently proceed.
Scope of these statements
These are high-level principles, not a description of every current control or a claim of perfect security. OSIO is in development; the public website’s guided simulation and previews are not an operational product security assurance. Product-specific security information should be stated separately when it is verified and applicable.