Diop Daily #146 — October 2026

Language Is the First Boundary

Every protocol contract describes identity, authorization, refusal, jurisdiction, and evidence. Most are written in English. The words are precise for a global audience, but the authority they describe is usually local: a university record governed by community conditions in Senegal, a public registry maintained in French and Arabic, a payment authorization subject to a national administrative system, a refusal that must be explained to a reviewer who reads in Wolof or Amharic. When the contract's authoritative language is foreign to the institution it governs, sovereignty does not fail loudly. It fails quietly — through misinterpretation, silent substitution, and refusal paths that produce technically correct but institutionally meaningless records.

A protocol that speaks only in the language of its authors governs only those who share that language. Everything else is administration by translation.

A field observation

I have been reading the new protocol announcements — Meta and Sierra's Personal Agent Protocol from October 6, the IETF agentproto charter from October 8, the W3C Agent Protocol Community Group's continuing work, and Google's A2A advances through ADK — not as a competitor or commentator, but as someone trying to understand where the authority lives. The signal is clear: the protocol layer is becoming the sovereignty surface. The detail that worries me is not the scope of these initiatives. It is the language they assume.

A protocol contract is not merely a specification of fields and types. It is the text that defines what refusal means, which jurisdiction applies, whose identity is valid, and what evidence must survive. If that text is only available in English, an institution that operates in another language cannot inspect its own obligations independently. A reviewer who reads French or Wolof cannot confirm that a refusal was issued correctly. A regulator who requires evidence in local form receives a record that is accurate in translation but not authoritative in the original context.

The cost of a translated contract

The Personal Agent Protocol's announcement describes discovery, authorization, session management, and secure interaction. These are serious structural elements. However, the public documentation released so far does not specify the language of authoritative refusal, the language of jurisdiction markers, or the requirement that evidence be preserved in the language of the institution rather than the language of the provider. That omission matters because a refusal expressed in English may satisfy a provider's audit while failing a local authority's inspection.

The IETF agentproto charter from October 8 defines user-to-agent, agent-to-agent, and agent-to-tool interactions with authorization and delegation as first-class concerns. The charter is well-scoped. Yet nothing in the public draft requires that the authorization policy, the refusal reason, and the delegation record be available in the language that the governing institution recognizes as authoritative. A university in Dakar that uses a French-language administrative system could adopt a protocol that technically supports delegation without being able to prove the delegation in its own language.

Google's A2A work through ADK provides multi-language SDKs — Python, Java, and Go — which demonstrates that portability is being treated as a runtime property rather than a framework accident. But runtime portability and linguistic authority are different things. A protocol that allows an agent to run in Java does not guarantee that the institution governing the agent can read its contract, inspect its refusal, or verify its jurisdiction marker in the language it uses for official records.

Where language becomes the boundary

The issue is not cosmetic translation. It is structural. Consider the fields a serious protocol contract should carry, and the language question each raises:

  • Identity contract: the agent's identity must be verifiable by the institution. If the identity document is in English and the institution's identity registry is in French, the verification process introduces an intermediary that the institution cannot audit.
  • Refusal contract: a refusal must explain why an action was denied, in a language the reviewing authority understands. A refusal expressed in English may be technically correct but institutionally useless for a regulator or community board that operates in another language.
  • Jurisdiction marker: the jurisdiction governing the interaction must be named in the language the institution uses for legal records. A jurisdiction label in English may describe the correct region but fail the institutional test of authority.
  • Evidence contract: the evidence preserved for each interaction must remain inspectable independently of the provider. If the evidence is available only in English, an institution that operates in French, Arabic, or Wolof cannot examine it directly.

The protocol is the first sovereignty surface, but language is the first sovereignty test. A contract an institution cannot read in its own language governs it by default.

Why African institutions cannot treat language as secondary

African institutions operate across multiple linguistic authorities by design. A national university may maintain records in French. A regional cooperative may govern membership in a local language. A public agency may serve citizens in several languages with different legal obligations attached. A publisher or archive may hold text in Wolof, Amharic, Arabic, or English with separate conditions for each edition. These are not edge cases. They are the operating environment.

If agent protocols treat English as the default authoritative language, African institutions that adopt them will inherit a governance structure they can use but not fully inspect. They can interact with the protocol, but they cannot independently confirm that refusal, delegation, jurisdiction, and evidence were handled in ways that respect their own linguistic authority. The dependency is not dramatic — it is quiet, cumulative, and difficult to reverse once adopted.

The durable asset in this environment is not the model or the endpoint. It is the institution's ability to state, in its own language, what authority permitted the interaction, what refusal occurred, what jurisdiction applied, and what evidence remains available. A protocol that supports that capability provides real sovereignty. A protocol that does not provides technical interoperability with governance by translation.

The standard is not finished

The good news is that the protocol layer is still open. The Personal Agent Protocol has only just announced. The IETF agentproto charter is at its beginning. The W3C group continues to develop contributions rather than finalize standards. There is time — not much, but enough — for institutions to insist that language be treated as a structural property rather than a cosmetic one.

This does not mean every standard must be translated into every language. It means the contract itself must specify which language is authoritative for identity, authorization, refusal, jurisdiction, and evidence. It means refusal records must be preserved in that language. It means jurisdiction markers must reference the language of governing authority. It means evidence must be available for inspection without requiring an intermediary to translate the institution's own obligations.

Where the investable surface is widening

The commercial object is not a language service added to a protocol. It is a protocol contract that treats language as authority. The categories that become strategic include:

  • Multilingual authorization contracts: machine-readable contracts that state identity, delegation, jurisdiction, refusal, and evidence in the language recognized by the governing institution.
  • Authoritative refusal records: refusal paths that produce evidence in the institution's language, with the reason, jurisdiction, and authority clearly named, not summarized after translation.
  • Jurisdiction-bound evidence storage: storage layers that preserve interaction records in the language of the governing authority, independent of the provider's preferred format.
  • Language-conformance evaluation: test environments that verify whether protocol implementations preserve institutional meaning across French, Arabic, Wolof, Amharic, Swahili, and English contexts.
  • Portable linguistic authority: contracts that can move between providers without changing the language of their governing obligations, so that a provider change does not change the authority under which the institution operates.

A practical test

Before adopting any agent protocol for an institution that operates in a non-English context, run a simple test. Request the interaction in the institution's authoritative language. Confirm that the refusal reason, jurisdiction marker, identity verification, and evidence record are available in that language without requiring translation by the provider. Confirm that a reviewer in that language can understand the record independently. Confirm that a change of provider does not change the language of the governing contract.

If any of those steps fails, the protocol provides interoperability but not sovereignty. It connects the institution to the agent ecosystem without giving the institution control over the terms of that connection. That is a practical result, not a philosophical objection. The institution that can verify these properties independently owns the relationship. The institution that cannot is administered by a contract it did not fully design.

A closing note

The agentic future will be described through endpoints, protocols, and contracts. African institutions will experience it through the languages they use to govern, review, correct, and recover. A protocol that treats language as decorative will produce systems that work smoothly until the first serious refusal, the first disputed jurisdiction, the first evidence request that an institution must explain in its own language. Building protocols that survive those moments requires making language structural from the beginning — not as a translation layer, but as part of the contract itself.

Sources