A Response to Anthropic's Statement on the Department of War
Dario Amodei is right about one thing: the architecture of AI matters enormously for democratic values.
Where he is wrong is in thinking Anthropic's architecture protects those values.
It does not. It never did.
Anthropic built a centralized system. Every query, every document, every prompt from every customer – flows through Anthropic's centralized servers. The data is processed, logged, and retained within the infrastructure they control. That was not accidental. Centralization improves model training, product iteration and customer leverage. It is an effective business model.
But it is not a privacy-preserving one.
Anthropic has built one of the most powerful data aggregation infrastructures in existence and positioned it as safety-focused. At the same time, its leadership warns about the risk of a surveillance state.
The contradiction is hard to ignore.
A centralized, cloud-dependent architecture where every interaction is logged and accessible to a private company— is the surveillance state. The Department of War didn't create that risk. Anthropic did. Willingly. Profitably.
At webAI, we made a fundamentally different choice.
Our systems operate on-device, at the edge, and in air-gapped environments. Inference can happen locally, in a private cloud, or across a hybrid topology — the choice belongs to our partners, not us. What doesn't change is custody: data never leaves the operator's control. We do not see customer inputs. We do not store them. We cannot provide access to data we do not possess. This is not a policy preference; it is an architectural constraint.
If mass domestic surveillance is incompatible with democratic values, then the solution is to design systems where large-scale centralized data capture is structurally impossible.
The Department of War is justified in seeking AI infrastructure it can rely on without conditional access or shifting corporate policy. When a private company publicly reserves the right to withhold capabilities based on its own judgement, it positions itself not merely as a vendor, but as a gatekeeper.
America does not need a gatekeeper.
It needs AI infrastructure that is sovereign by design — distributed, on-device, architecturally private, and unambiguously committed to the warfighter.
webAI is built for that future.
David Stout
CEO, webAI