Oracle Expands Fusion Claw with Outcome Receipts and an Enterprise Operating Envelope

At Oracle AI World on 26 October, Oracle expanded Fusion Claw, its governed execution runtime for agentic applications, first introduced on 29 September. The architecture separates three layers: the LLM reasoning layer, an Enterprise Operating Envelope defining authority such as goals, SOPs, policies, permissions, risk thresholds, decision rights and escalation points, and deterministic execution in an isolated environment guarded by an Outcome Trust Harness. Crucially, the LLM cannot grant itself additional permissions because the authority layer is separated from reasoning.

Every execution ends with an Outcome Receipt: an auditable record of the authority used, evidence consulted, decisions taken, transactions executed and the result. In a keynote demo, an agent analysed 154 logistics shipments, proposed consolidating two with an estimated $12,000 saving, escalated five to humans, and ran the remaining 149 automatically. Oracle also claims customers pay tokens only for reasoning, not for execution on its compute layer. The demonstration figures are vendor claims.