Oracle has introduced Fusion Claw, a governed agentic execution runtime for its Fusion Agentic Applications that combines AI reasoning with deterministic enterprise computation, aimed at letting complex work run more autonomously and at scale.

As AI shifts from assisting people to executing tasks on its own, organisations need control over what agents can access, decide and change. Fusion Claw applies what Oracle calls an Enterprise Operating Envelope, which sets an organisation’s objectives, permissions, risk thresholds, approval requirements and escalation boundaries before an agent starts work.

Oracle has also introduced 25 Claw-powered Fusion Agentic Applications, bringing its total portfolio to 75. The new applications span finance, HR, supply chain and sales — including reconciling accounting records, developing staffing plans, consolidating shipments and planning sales territories.

Governance built into every outcome

Each outcome run carries what Oracle calls an Outcome Trust Harness, applying specific authority and controls covering identity, capabilities, data and actions. An accompanying Outcome Receipt records the authority applied, the evidence used, the decisions made and the actions and transactions executed, alongside the result — creating an audit trail for each piece of completed work.

Oracle said a frontier model handles the reasoning, planning and adaptation, while deterministic computation takes over for high-volume execution, an approach it argues gives customers an economic advantage over running everything through a language model. Customers can also choose the level of automation for each application — whether it assists, recommends a course of action, or completes approved work independently.

Fusion Claw runs on Oracle Cloud Infrastructure and is powered by frontier models including Gemini and OpenAI, with support for additional models planned over time.