EU AI Act Weekly Radar: transparency moves from preparation to enforcement

This week marks a practical turning point for the EU AI Act. The European Commission says that from 2 August 2026, the AI Office and national authorities begin enforcing the AI Act rules that are already applicable, alongside new transparency requirements. In parallel, the Commission says around 190 organisations signed the Code of Practice on Transparency of AI-generated Content before those legal obligations took effect.

Taken together, these updates suggest that the current phase of implementation is no longer mainly about future-readiness. For many providers and deployers, it is now about whether existing product flows, disclosures, and escalation channels are operational today.

What changed this week

In its 31 July announcement, the European Commission said it would start enforcing applicable AI Act rules and new transparency requirements on 2 August. The Commission highlighted duties relating to chatbots, deepfakes, and machine-readable marking of AI-generated or altered content. It also pointed stakeholders to complaint, whistleblower, and downstream-provider channels in connection with enforcement and reporting pathways. These points are set out in the Commission news item, "Commission starts enforcing AI Act rules and new transparency requirements on 2 August".

Also on 31 July, the Commission reported strong uptake for the transparency code, saying that about 190 organisations had signed the Code of Practice on Transparency of AI-generated Content before the legal obligations took effect. According to the Commission, the code offers a streamlined way to demonstrate compliance for providers and deployers of generative AI systems, and signatories include both major AI vendors and many smaller companies. That update appears in the Commission news item, "Strong backing for the Code of Practice on Transparency of AI-generated Content".

Why these two updates matter together

The two announcements reinforce the same message from different angles.

First, the enforcement notice indicates that the Commission now expects covered organisations to have practical compliance measures in place for the transparency obligations that have become applicable.

Second, the strong sign-up numbers for the transparency code indicate that many market participants are not waiting to build bespoke evidence frameworks from scratch. Instead, they appear to be using a Commission-backed code as a more standardised route to show alignment.

For lextrace readers, that combination matters because it changes the compliance posture from interpretation only to implementation plus defensibility. Once enforcement starts, the question is no longer just what the rule means in theory. It becomes:

• what the user sees,
• what metadata or machine-readable markers are attached to content,
• what internal teams do when concerns are raised,
• and what evidence an organisation can produce if challenged.

The immediate regulatory significance

1. Transparency is now an operational issue

The Commission specifically highlighted transparency duties tied to chatbots, deepfakes, and machine-readable marking of AI-generated or altered content. Even without additional detail in this week’s source items, that is enough to show that transparency is not being framed as a documentation-only obligation. It is being framed as something that must appear in products, outputs, or related technical controls.

That has direct implications for teams working across:

• product UX,
• model output handling,
• trust and safety,
• content pipelines,
• platform policy,
• incident intake,
• and legal/compliance review.

If a system generates or alters content in ways covered by the applicable rules, the visibility and format of those disclosures now matter in an enforcement context.

2. The code of practice is becoming a market signal

The Commission describes the transparency code as a streamlined route to demonstrate compliance for providers and deployers of generative AI systems. That does not mean the code replaces the law. But it does suggest that the Commission sees the code as a practical compliance mechanism rather than a purely symbolic initiative.

The reported participation level, around 190 organisations, is significant for another reason: it may help create market expectations around what “good enough” transparency implementation looks like. Large vendors and smaller companies signing the same code can push toward more consistent approaches in areas such as disclosure language, marking methods, and internal governance processes.

For startups and SMEs, that may reduce some uncertainty compared with designing a standalone transparency framework without any recognised benchmark.

3. Enforcement channels are part of the story

The Commission’s enforcement announcement did more than say that rules are live. It also pointed to complaint, whistleblower, and downstream-provider channels.

That matters because enforcement risk does not arise only from proactive regulator review. It can also arise from:

• customer complaints,
• partner concerns,
• internal reports,
• and issues escalated by other actors in the value chain.

In practical governance terms, that means organisations should not treat transparency compliance as isolated from reporting and remediation. A disclosure failure, missing marker, or unclear product flow can quickly become an escalation issue once reporting channels are actively promoted by the Commission.

What providers and deployers should be watching now

Based on this week’s official updates, several themes stand out.

User-facing disclosures

If your tool includes conversational interfaces or generates content that users may mistake as purely human-created or unaltered, the Commission’s focus on chatbot and deepfake transparency means disclosure design deserves immediate scrutiny. Teams should be asking whether disclosures are present, understandable, and consistently triggered in the relevant use cases.

Machine-readable marking

The Commission explicitly referenced machine-readable marking of AI-generated or altered content. That elevates the importance of technical implementation choices, not just front-end text. Organisations should know where such marking is applied, how consistently it persists through workflows, and which product or engineering owners are accountable for it.

Role clarity between providers and deployers

The Commission’s description of the transparency code says it offers a streamlined way to demonstrate compliance for both providers and deployers of generative AI systems. That is important because many organisations occupy more than one role across their stack. A company may build, fine-tune, integrate, and deploy AI features in different combinations.

This week’s updates therefore strengthen the case for mapping obligations by role rather than assuming compliance sits entirely with the original model vendor.

Intake and response channels

Because the Commission pointed to complaint, whistleblower, and downstream-provider channels, organisations should understand their own internal routing. If an issue is raised externally, can the company:

• identify the affected system,
• trace the relevant disclosure logic,
• confirm whether marking was applied,
• and document what remediation was taken?

Those questions sit at the intersection of legal, product, and operations.

What this means for SMEs and startups

The transparency code’s uptake by both major vendors and smaller companies is one of the most notable signals in this week’s news.

For smaller teams, the Commission’s framing suggests a potentially useful path: align with a recognised code where possible, rather than trying to independently justify every design and governance choice from first principles. That may help startups show investors, enterprise customers, and internal stakeholders that their transparency approach is grounded in an EU-backed framework.

At the same time, this week’s enforcement message means smaller size is not the same thing as delayed urgency. If applicable transparency obligations are in force and enforcement has started, product teams still need working controls. The code may streamline demonstration, but it does not remove the need to implement.

The broader implementation signal

Although this week’s source items focus on transparency and enforcement rather than the full architecture of the EU AI Act, they still tell us something important about the implementation timeline.

The Commission is pairing:

• a clear public enforcement start date,
• public messaging around specific live obligations,
• and a practical compliance instrument with strong reported adoption.

That is a classic sign of regulatory transition from awareness-building to supervision.

For lextrace readers tracking the wider EU AI Act roadmap, the key takeaway is not that every question is settled. It is that some obligations are now being treated as mature enough for active enforcement, and the Commission is openly signalling both the expected behaviours and the available reporting pathways.

lextrace takeaways

This week’s radar can be distilled into three points:

1. 2 August 2026 is an enforcement milestone. The Commission says the AI Office and national authorities begin enforcing applicable AI Act rules and new transparency requirements from that date.
2. Transparency is where implementation becomes visible. Chatbot disclosures, deepfake-related transparency, and machine-readable marking are not abstract governance topics; they are observable product and workflow decisions.
3. The transparency code is gaining real compliance weight. With about 190 organisations reportedly signed on before the obligations took effect, the code is emerging as a meaningful benchmark for how providers and deployers may choose to evidence alignment.

For organisations still treating transparency as a near-term planning issue, this week’s Commission updates suggest the window for preparation has narrowed. The practical question now is whether your systems, controls, and escalation routes are ready to withstand scrutiny in an enforcement environment.