HypothesisVery high confidenceCurrent
Consequential authorization and execution must be enforced outside the language model.
The Dataheart model runtime is a proposal engine.
Valid from2026-08-23
Last reviewed2026-08-23
Review again2026-11-21
Dataheart stanceexploring
Evidence for
OAuth and conventional authorization rely on deterministic resource-server enforcement.
RFC 9700: Best Current Practice for OAuth 2.0 Security · rfc-editor.org ↗MCP Tools specification · modelcontextprotocol.io ↗Defeating Prompt Injections by Design, CaMeL · arxiv.org ↗MCP requires servers to implement access control and input validation rather than trusting model behavior.
RFC 9700: Best Current Practice for OAuth 2.0 Security · rfc-editor.org ↗MCP Tools specification · modelcontextprotocol.io ↗Defeating Prompt Injections by Design, CaMeL · arxiv.org ↗CaMeL obtains scoped guarantees by enforcing policy in a separate interpreter and capability layer.
RFC 9700: Best Current Practice for OAuth 2.0 Security · rfc-editor.org ↗MCP Tools specification · modelcontextprotocol.io ↗Defeating Prompt Injections by Design, CaMeL · arxiv.org ↗Evidence against
External enforcement adds latency, implementation effort, and policy-authoring burden.
RFC 9700: Best Current Practice for OAuth 2.0 Security · rfc-editor.org ↗MCP Tools specification · modelcontextprotocol.io ↗Defeating Prompt Injections by Design, CaMeL · arxiv.org ↗Some low-risk read-only use cases may accept lighter controls.
RFC 9700: Best Current Practice for OAuth 2.0 Security · rfc-editor.org ↗MCP Tools specification · modelcontextprotocol.io ↗Defeating Prompt Injections by Design, CaMeL · arxiv.org ↗How this claim can fall
Replace this placeholder during human review with an observable result that would materially weaken or falsify the claim.
A published claim remains useful only while it remains possible to prove it wrong.