Brevity codes: what they are and why a system must not guess one

In radio and chat traffic across the Services, a brevity code stands in for a longer message. A decision-support tool that meets a code it does not hold must say so, because a code's meaning comes from an agreement among users and not from the word itself, and a meaning the tool invents is a fabricated fact that a person may act on. Brevity codes also add no security: the publisher of the multi-Service list says they only shorten transmissions.

What a brevity code is

The Air Land Sea Space Application (ALSSA) Center publishes multi-Service tactics, techniques, and procedures. The center's Brevity page states that a brevity code provides no additional communications security and that brevity codes only serve to shorten transmissions. A brevity code is an abbreviation by agreement, not a cipher.

The Marine Corps library listing for MCRP 3-30B.1 gives the scope of the same publication. It standardizes air-to-air, air-to-surface, surface-to-air, surface-to-surface, maritime air, and maritime-to-maritime codes, whether the medium is voice or chat, and it covers abbreviations specific to tactical chat.

One publication, five Service numbers

The full title is Multi-Service Tactics, Techniques, and Procedures for Multi-Service Brevity Codes. One document carries a different number in each Service: ATP 1-02.1 for the Army, MCRP 3-30B.1 for the Marine Corps, NTTP 6-02.1 for the Navy, AFTTP 3-2.5 for the Air Force, and STTP 3-9002 for the Space Force. ALSSA says the publication defines and standardizes the multi-Service brevity codes agreed upon by each U.S. Service branch. The ALSSA page for the April 2025 edition carries a Public Release label, and the Marine Corps library lists MCRP 3-30B.1 as dated April 25, 2025 and current.

ALSSA makes three statements about the reach of the publication:

• It does not include Service-specific brevity codes.
• It is not synonymous with NATO's APP-7.
• Updates have been shared with the NATO Standardization Office for inclusion or modification into Allied Communications Publications.

The ALSSA page also carries a history table. It lists editions published in May 2020, March 2023, and April 2025, a Change 1 published in April 2021, and, above the 2025 entry, a joint working group in May 2026 and a worldwide review in August 2026. The set of codes is maintained over time, and any copy of it is a snapshot.

Why a system must not guess

Picture a decision-support tool that answers questions over a library of documents and meets a code word that the library does not contain. Three points show why any meaning it composes is unsupported.

The meaning is assigned, not derived. A code word is defined by agreement. Spelling and surrounding words do not determine what it stands for, so there is nothing for a model to work out. It can only retrieve the definition or invent one.

Absence from one publication is not absence of meaning. The word may be a Service-specific code, which the multi-Service publication excludes. It may be an Allied code, and ALSSA states the publication is not the same as APP-7. It may have been added or modified in an edition the tool does not hold. In each case the definition sits in a source the tool does not have, and a guess cannot find it.

Language models lean toward answering. In a September 2025 paper, Kalai and colleagues argue that models hallucinate because training and evaluation procedures reward guessing over admitting uncertainty. A survey in ACM Computing Surveys describes sequence-to-sequence models as prone to hallucinate unintended text. Both papers describe general tendencies that a designer has to account for.

What refusing looks like

The Department of Defense (DoD) adopted AI ethical principles on February 24, 2020. The principles are stated for AI capabilities in general, and three of the five apply to a code lookup:

• Reliable: capabilities have explicit, well-defined uses and are tested within those defined uses.
• Traceable: relevant personnel understand the technology, with transparent and auditable methodologies, data sources, and documentation.
• Governable: capabilities can detect and avoid unintended consequences.

A tool whose defined use is looking up codes in named publications can be tested for that use. Composing a meaning for a missing code is a different use, outside what that testing would cover.

Refusal then has a concrete form. The tool states that the term is not in the sources it holds, names the publications and editions it searched, and returns the question to the person. That answers the question the user asked, which is whether the library defines the term. A guess answers a different question, what the word might mean, without saying so.

What to ask

• When a code is not in the library, does the tool say so and return the question to the person?
• Does the answer name the publications and editions the tool searched?
• Does the tool separate multi-Service codes from Service-specific codes and Allied codes?
• Which edition does the tool hold, given that the publication has been revised several times, with editions in May 2020, March 2023, and April 2025?
• Can every definition the tool returns be traced to the publication it came from?

Azcuna's ELLMENT, the E-2D Large Language Model Entity, was proposed to NAVAIR PMA-231 under Navy SBIR Phase I topic DON26BZ01-NV010. The prototype, built at private expense on publicly available doctrine, answers with a page-level citation and refuses when the documents do not support an answer. The proposal has been submitted. The long read on air-gapped RAG goes further.

Sources

1. 1Brevity · Air Land Sea Space Application (ALSSA) Center
2. 2MCRP 3-30B.1 · United States Marine Corps, Marine Corps Publications Electronic Library
3. 3DOD Adopts Ethical Principles for Artificial Intelligence · U.S. Department of Defense (February 24, 2020), now hosted on war.gov
4. 4Why Language Models Hallucinate · arXiv (Kalai, Nachum, Vempala, Zhang)
5. 5Survey of Hallucination in Natural Language Generation · arXiv / ACM Computing Surveys (Ji et al.)