Deliverance AI says it can run enterprise AI agents inside customer-controlled infrastructure, with controls for agent behaviour and model choice. The pitch is aimed at regulated businesses that need AI workflows without handing over the keys to their operating environment.
Watch Desk analysis
What happened
In a SiliconANGLE interview, Deliverance chief executive Mick McNeil described a control plane that observes agent activity, enforces policies, keeps audit trails and includes a kill switch. The company says it works with HPE to deploy a full stack of storage, networking, GPUs and software in customers’ data centres. Some deployments may be air-gapped; others can allow selected external connections.
McNeil says Deliverance’s routing technology can direct workflows to frontier or open-source models, including a fallback if a provider restricts access. Customers retain their agents, integrations, audit trails and accumulated learning, he says, though moving those assets between platforms remains a separate challenge. Deliverance also says a joint HPE solution for deploying its agentic operating system is due to launch within 90 days. Those are company claims and plans described in the interview, not independently demonstrated results. SiliconANGLE disclosed that the interview series was sponsored by HPE.
Why it matters
For organisations with strict data or operational requirements, sovereignty is about more than where information is stored. It also means deciding what agents can do, which services they depend on and what happens if a model provider becomes unavailable. A customer-hosted control layer with model routing and a fallback option addresses those practical concerns more directly than a flag on a data-centre map.
The trade-off is that customer control does not automatically mean portability. Deliverance says customers keep their agents and related assets, but acknowledges there is no established framework for moving them between platforms. The price, access terms and independent evidence of performance are also not supplied in the interview.
Our read
This is a substantive enterprise AI proposition: keep infrastructure under the customer’s control, govern agents at runtime and preserve a route to alternative models. The useful test is not whether the sovereignty language sounds reassuring, but whether the controls work across real workflows and whether customers can leave without rebuilding everything from scratch.
What to watch
- Whether the announced HPE joint solution launches on the stated timetable.
- Which models, external connections and deployment options customers can actually use.
- How customers can export or transfer agents, integrations and audit trails.
- Evidence on performance, costs and the effectiveness of the control plane in production.
Discussion spark: For regulated organisations, is keeping AI agents inside customer-controlled infrastructure worth the extra complexity if moving them between platforms remains difficult?
Sources and evidence
- Deliverance AI puts sovereign AI control to the continuity test – SiliconANGLE (13 October 2026, 19:34 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.