Most of the writing about the EU AI Act and agents describes a problem and stops there. "The EU AI Act's runtime problem." "Agents now have to say who sent them." The obligation is named; the owner of the answer is absent. This post takes the next step: what the obligation actually is, who carries which duty under Article 50, and the named standard for a runtime record that proves enforcement happened.

The short version: agents that deal with people must disclose their artificial nature and the person on whose behalf they act (Article 50(1) and the Commission's guidelines frame that as a design-and-disclose duty for providers). Deployers have their own Article 50 duties under 50(3) and 50(4). Separately, organisations that want a regulator-readable trail of what the runtime decided need a record. That record problem is what the Cited Decision Standard names.

What Article 50 actually requires

Article 50(1) of Regulation (EU) 2024/1689 requires providers of AI systems that interact directly with natural persons to ensure those persons are informed they are interacting with an AI system, unless it is obvious. Article 50(3) and 50(4) add deployer duties: notice for emotion recognition and biometric categorisation, and disclosure of deepfakes and AI-generated text of public interest. These transparency obligations begin to bind on 2 August 2026, the same date the penalty powers for general-purpose AI providers take effect.

One boundary matters and is easy to overstate. Pure machine-to-machine traffic, one service calling another, is outside the scope of Article 50(1). The obligation attaches when an AI system deals with a person. So the defensible line is not that the agent economy cannot transact. It is narrower and sharper: an agent that deals with a person has to declare what it is, and on whose behalf it is acting.

The load-bearing text: the Commission's Article 50 guidelines

On 8 May 2026 the Commission's AI Office published draft guidelines on the implementation of the Article 50 transparency obligations. The Commission adopted the final guidelines on 20 July 2026. They apply alongside Article 50 from 2 August 2026, and they are where the agent question is answered directly.

The guidelines say that agents capable of making bookings, managing correspondence, negotiating or concluding contracts, or executing purchases must be designed and developed to disclose both their artificial nature and the person on whose behalf they act. The stated reason is precise, and it is the whole argument: to make the origin of the interaction, the delegation of authority, and accountability for the resulting actions visible.

Read that against how agents are actually being shipped. An agent books a flight, sends an email, agrees to terms, completes a purchase. The counterparty is a person, or a company acting through people. The Act expects that interaction to carry a clear disclosure of artificial nature and delegated authority. That is a transparency design duty, not a free-standing statute that invents a deployer log format. Producing a reviewable runtime trail of what was allowed or blocked is a separate operational problem, and it is the one the Cited Decision Standard addresses.

The Code of Practice asks for compliance documentation. It does not produce the runtime record.

The Commission published the Article 50 Code of Practice on 10 June 2026. It is voluntary implementation support, not a new obligation. Article 50 of the Regulation remains the legal baseline. For deployer signatories, Measure 2.1 commits them to internal compliance processes and documentation of how disclosure obligations are implemented, not a statutory artefact named an "evidence record."

Here is the gap that matters. The Code assumes you can produce that documentation. It does not supply machinery that writes a decision record at the moment an AI system acts. The tooling that has grown up around the Code is browser-side checklists and marking validators that, in their own words, do not make a legal or compliance determination. Separately, where a deployment is high-risk, Article 26(6) sets a six-month floor for automatically generated logs. That retention duty is not an Article 50 inventing a general record-keeping clause for every deployer.

So the deployer is left with a Code commitment to document disclosure practices, a product need for a defensible runtime trail, and nothing in the free tooling box that actually writes the record when the AI makes the call.

The Cited Decision Standard

Call the required record what it is. The Cited Decision Standard is a named, citable rule for EU AI Act runtime enforcement audit trail evidence, built from artifacts that already ship (the open-source cited Rego corpus, the Article 12 audit export, and the published GPAI benchmark), not from a new product layer.

Under the Cited Decision Standard, every logged decision carries:

When a rule fires and the decision is a denial, the record also carries an article citation at article, paragraph, and sub-paragraph where the rule states them (verbatim legal basis on the decision, not a generic "flagged" label). Allow-path events keep hash, timestamp, and chain without inventing a citation that was never applied.

ComplyEdge sits between the model and the user and enforces the Regulation at runtime, on every request. The block carries the law: an article-cited denial, plus regulator-readable evidence, per decision. Not a probability score, and not a log line that says "flagged" with no article attached. Export fields map to the standard: text_hash, timestamp, and chain_link on each event; when a rule fires, violations[].rule_description carries the article citation; /v1/audit/export is the regulator-readable package. ComplyEdge retains the trail for 180 days (a product guarantee; the Article 26(6) six-month floor binds only for high-risk deployments from 2 December 2027).

The enforcement corpus is open source: more than fifty deterministic policies written in Rego, covering Article 5 prohibited practices, Article 50 transparency, the high-risk requirements, and the general-purpose AI chapter, each carrying a verbatim citation. On the published runtime benchmark, the core EU AI Act categories detect at 100 percent with zero false positives on the safe-harbour set; the benchmark code and result JSON are in the repository, dated, and reproducible against the live API with any key. A durable, versioned home for the Cited Decision Standard (stable URL, semver, citation block) is a separate publication; this post is the manifesto that names the standard and shows how the existing artifacts already satisfy it.

Say what this is, and what it is not

ComplyEdge is the enforcement and evidence layer that implements the Cited Decision Standard at runtime. It blocks a violation with the article, and it produces a tamper-evident record of every decision (with an article citation when a rule fires). That is a real and, as far as we can find, unoccupied position: the AI-governance vendor landscape is full of tools that log activity, and short of tools that hand a regulator a cited legal basis for each blocked decision.

It is worth being equally clear about what ComplyEdge is not. It is not an agent identity system, and it does not issue or verify the mandate that authorizes an agent to act. Those are separate problems, and other layers will own them. What ComplyEdge does is enforce the Regulation at the point of action and produce the Cited Decision Standard record that proves the enforcement happened. That claim is narrow, and it is verifiable, and it is enough.

Where this sits in the stack

The stack a compliance-conscious AI team is being taught to assemble has a governance and risk layer, a system of record, and an observability layer. Those tell you what your AI did, after the fact. None of them enforces a regulation at the instant the AI acts, or hands a regulator the cited proof that it did.

That is a fourth box: a runtime control plane to enforce and prove. It is where the article-cited denial and the tamper-evident record come from. When an agent acts for a person under Article 50, that fourth box is what turns "we have logs" into "here is the record, and here is the article it satisfies."

The clock, stated honestly

Two dates, kept straight. 2 August 2026: Article 50 transparency obligations bind, and the penalty powers for general-purpose AI providers take effect. The high-risk regime under Annex III did not move up; the Digital Omnibus deferred it to 2 December 2027. This is a readiness argument, not a panic one. The transparency duty is live now; the record-keeping capability is the thing worth having in place before a regulator, a counterparty, or a law firm asks to see it.

You can read the enforcement corpus, the Python SDK, and the runtime benchmark, all open source under Apache 2.0:

pip install complyedge

Repository: github.com/ComplyEdge/complyedge
Deployer obligations, mapped to the mechanism: Article 26 one-pager
Quick start: docs/quick-start