EOS Intelligence
Governed SI: agents propose, people decide
EOS Intelligence reads your reconciled figures and the records behind them, explains differences and proposes what to do. It does not decide or change records. A named person decides, and that decision is recorded.
A short path through this page. Open each step to go deeper.
1The rules, as built
- An agent cannot decide. An agent may research, correlate, explain, simulate and propose. A named person decides a proposal under the organization's policy.
- An agent cannot act on its own. A change is carried out only for a request a person approved, under a signed, single-use authorization, and then verified.
- Keys die with their maker. An agent key stops working when the person who created it leaves.
- Authority is per person and per legal entity. A person decides for an entity only when it is recorded in their scope.
- No self-approval. The person who raised a request cannot approve it, and large amounts need two people.
2Every figure says what it rests on
- OBSERVED
- Directly supported by a dated record.
- DERIVED
- Calculated from observed figures; the calculation is shown.
- INFERRED
- Reasoned, not observed.
- SCENARIO
- A modeled future. Nothing was changed.
- DECISION
- A determination a person with authority made.
- ACTION
- Something actually carried out.
- OUTCOME
- What was measured afterward.
The SI does not present one as another: a recommendation is never reported as a decision, and a scenario is never reported as an action.
3Tested against attack, not only against mistakes
Text from outside the organization, such as a document, a web page or a record, is treated as untrusted data, never as an instruction. A red-team suite tries eight attacks, including instructions hidden in records, attempts to approve one's own request and attempts to cross organizations, and each is checked by breaking the defense and confirming the test fails.