OpenTelemetry
A vendor-neutral standard for traces, metrics and logs, with a common data model and wire format. The dominant instrumentation standard in distributed systems.
Why it matters for autonomous finance
The most useful standard on this list, and the one least discussed in AI governance. An agent decision is a distributed trace — a chain of tool calls, retrievals and model invocations with causal relationships and timing. OpenTelemetry already has the data model for that, plus a decade of production tooling. Firms building bespoke 'AI audit logs' are rebuilding a solved problem badly, and ending up with something no auditor recognises.
What it does not cover
A trace records what happened, not what the model was. Reconstructing a decision requires model version, weights or endpoint identity, prompt, retrieval corpus state, and sampling parameters — none of which OpenTelemetry captures unless you deliberately attach them as span attributes. Instrumenting without that discipline produces logs that look rigorous and cannot reconstruct anything.
Carries no semantics for model identity, data provenance, or authority. Every entry in this register carries this section. A standard read past its scope is worse than no standard, because it produces confidence without coverage.