The World's First Real AI Deadline Was About Labels, Not Decisions

For nearly two years, "August 2026" functioned as the industry's shorthand for the moment artificial intelligence became genuinely regulated. Board packs were built around it. Budgets were defended with it. Vendors sold against it.

The date arrived on 2 August 2026. What landed was not what most organisations had prepared for.

Two things happened simultaneously, in two of the world's largest markets. In the European Union, the AI Act's transparency duties under Article 50 became applicable, and the European Commission's AI Office began enforcing the Regulation alongside national authorities (European Commission, 2026a). In California, the AI Transparency Act — enacted as SB 942 and amended by AB 853 — became operative on precisely the same day, an alignment its drafters chose deliberately (California Legislature, 2025).

Both laws regulate disclosure: telling people they are talking to a machine, and marking content a machine produced.

What did not happen is at least as important. Six days earlier, on 27 July 2026, Regulation (EU) 2026/1744 — the "AI Omnibus" — entered into force and moved the AI Act's high-risk obligations out of August 2026 entirely. Rules for Annex III systems, which govern AI used in recruitment, credit scoring, education, biometrics, migration and critical infrastructure, now apply from 2 December 2027. Rules for AI embedded in regulated products under Annex I apply from 2 August 2028 (European Commission, 2026b).

So the world's first hard AI compliance deadline governs what AI systems must say about themselves. The rules governing what AI systems may decide about people were postponed by sixteen to twenty-four months.

This article argues that the sequencing was defensible but is being widely misread — and that the gap between the two layers is now the single most useful thing for a professional to understand about AI governance. It is written for compliance leads, product managers, engineering directors, general counsel, risk officers and executives who must decide what to do with an unexpected extension.

---

Executive Summary

• On 2 August 2026, EU AI Act Article 50 transparency obligations became applicable and the AI Office began enforcement; California's AI Transparency Act became operative the same day.
• Regulation (EU) 2026/1744 (the AI Omnibus), in force 27 July 2026, moved Annex III high-risk obligations to 2 December 2027 and Annex I to 2 August 2028.
• The stated rationale is readiness: harmonised standards under CEN-CENELEC JTC 21 slipped repeatedly, and without them Article 40's presumption of conformity has nothing to attach to.
• Disclosure rules shipped first because they are the only tier that is legally operable without a completed standards stack.
• The technical foundation for content marking is weaker than its legal status implies. Provenance metadata can be stripped; invisible watermarks are removable under published regeneration attacks (Zhao et al., 2024).
• The Commission's own approach concedes this: its voluntary Code of Practice combines metadata, watermarking and logging rather than relying on any single mechanism.
• The Omnibus was not purely deregulatory — it added a ninth prohibited practice covering AI-generated non-consensual intimate imagery and CSAM, effective December 2026.
• Article 50 breaches sit in the same penalty tier as high-risk breaches: up to €15 million or 3% of worldwide annual turnover, whichever is higher (Art. 99(4)(g)).
• The principal risk to organisations is substitution: treating a visible disclosure banner as evidence of governance maturity it does not represent.
• The delay is best understood as engineering runway with a shortened standards window, not as relief.

---

Why the Labelling Layer Shipped First

The explanation is structural, not political, and it is worth understanding precisely because it predicts what happens next.

The AI Act's high-risk regime is a product-safety regime. It works the way EU rules for lifts, toys and medical devices work: the law states outcomes, and harmonised technical standards — cited in the Official Journal — define how to demonstrate you achieved them. Under Article 40, conformity with a cited harmonised standard creates a legal presumption of conformity with the corresponding requirement. That presumption is the mechanism that makes the regime workable for engineering teams.

Those standards were not ready. CEN-CENELEC's Joint Technical Committee 21 missed its original April 2025 delivery target, and in October 2025 the CEN and CENELEC Technical Boards adopted an exceptional acceleration package — including direct publication after a positive Enquiry vote, bypassing separate Formal Votes — specifically to get priority deliverables available by Q4 2026 (CEN-CENELEC, 2025). The measures themselves drew criticism from within the committee, on the grounds that emergency procedure displaces the consensus-building that gives European standards their legitimacy (Cantero Gamito, 2025).

Strip away the procedural detail and the position in mid-2026 was this: a legal regime designed to be operationalised through standards, arriving before the standards existed.

Article 50 has no such dependency. "Inform the user that they are interacting with an AI system" and "mark synthetic content in a machine-readable format" are duties a competent team can implement from the statutory text. They needed guidance, not a conformity infrastructure — and guidance is what the Commission produced, publishing transparency guidelines on 20 July 2026 and a voluntary Code of Practice on transparency of AI-generated content, which more than 180 organisations had signed by the enforcement date (European Commission, 2026a; European Commission, 2026c).

Disclosure went first because disclosure was the only tier that could.

Figure 1 — Diagram brief for the OneWise design team

Title: Two Layers, Two Clocks: How the AI Act Regime Split Purpose: Show why one compliance tier became operable in 2026 and the other did not. Structure: Two horizontal tracks running left to right along a shared timeline (Aug 2024 → Aug 2028). Upper track — "Disclosure layer": Nodes: Statutory text (Art. 50) → Commission guidelines, 20 Jul 2026 → Voluntary Code of Practice → APPLICABLE, 2 Aug 2026 (solid green terminal node). Lower track — "Accountability layer": Nodes: Statutory text (Annex III, Arts. 9–15, 17) → Standardisation request M/593 + M/613 → CEN-CENELEC JTC 21 drafting → Citation in Official Journal (pending) → Art. 40 presumption of conformity → APPLICABLE, 2 Dec 2027 (amber terminal node). Key visual device: A dashed vertical line at 2 Aug 2026 crossing both tracks. It intersects a completed node on the upper track and an incomplete node on the lower track. Annotation callout: "The lower track has one dependency the upper track does not: published standards." Caption: The two tiers did not diverge for political reasons. They diverged because only one of them required a completed standards stack to function.

---

The Uncomfortable Part: The Labels Are Weaker Than the Law

Here is where a commentary has to say something a compliance memo will not.

Article 50 requires that AI-generated or manipulated content carry marks that are machine-readable and detectable as artificial. The dominant implementation combines two mechanisms: cryptographically signed provenance metadata, most prominently the C2PA specification's Content Credentials, and imperceptible in-content watermarking such as Google's SynthID.

Both have well-documented failure modes, and neither is a fringe concern.

Signed manifests travel with the file. Re-encoding, format conversion, upload pipelines that discard metadata, and screenshots all break the binding — and a screenshot is not an exotic adversarial technique. Invisible watermarks are meant to survive those transformations, but peer-reviewed work has shown that pixel-level invisible watermarks can be removed by regeneration attacks that add noise and reconstruct the image with a diffusion model, with formal guarantees and empirical results across multiple watermarking schemes (Zhao et al., 2024). Related work has argued that robust image detection faces fundamental limits against adaptive adversaries (Saberi et al., 2024), and that watermarking schemes strong enough to be undetectable rest on cryptographic secrecy assumptions (Christ et al., 2024). In text, watermarking research since Kirchenbauer et al. (2023) has repeatedly found that paraphrasing degrades detectability.

The Commission has not ignored this. Its Code of Practice takes a layered approach — metadata, watermarking and logging together — which is an implicit acknowledgement that no single mechanism carries the load.

But layering mitigates the wrong problem for most professionals. The operational hazard is not that a determined adversary defeats a watermark. It is the inference error: treating absence of a credential as evidence of authenticity. Unmarked content includes everything produced before provenance systems existed, everything from non-participating tools, everything that passed through a metadata-stripping platform — and everything deliberately laundered. A verification workflow that reads "no credential found" as "human-made" is worse than no workflow, because it manufactures false confidence at scale.

This is why we would resist the framing, common in vendor material this summer, that provenance "solves" synthetic media. It does something narrower and genuinely valuable: it lets a cooperating publisher make a checkable positive claim about its own content. That is a real capability. It is not a detector.

---

Table 1 — What Actually Changed, and When

Compiled by OneWise from Regulation (EU) 2024/1689, Regulation (EU) 2026/1744 and European Commission implementation pages, as at 11 August 2026. Confirm against the Official Journal before relying on any date.

---

Table 2 — Two Laws, One Date, Different Logic

Original OneWise comparison. Not legal advice; scope determinations are fact-specific.

---

Table 3 — Disclosure Mechanisms and Their Honest Limits

Original OneWise synthesis of published technical literature and standards documentation.

---

The Steelman: Why Disclosure-First Was Probably Right

An honest commentary has to argue against itself.

There is a serious case that the sequencing was correct rather than merely unavoidable.

First, enforcing high-risk obligations without cited harmonised standards would have produced arbitrary outcomes. Two firms with identical CV-screening systems could have been assessed differently by two national authorities, each improvising a benchmark. That is worse for fundamental rights than a delay, because inconsistent enforcement teaches the market that compliance is a lottery rather than a specification.

Second, the disclosure tier builds institutional capacity that the harder tier will need. Enforcement machinery does not appear on a start date. The AI Office, national market surveillance authorities, complaint channels and whistleblower tooling all now have a live, comparatively tractable obligation to practise on before the December 2027 wave arrives.

Third, disclosure has unusually broad reach for its cost. Article 50 touches nearly every organisation running a customer-facing assistant or publishing generated media — a far larger population than the Annex III set — and its marginal implementation cost is low.

Fourth, the Omnibus was not a retreat across the board. It added a prohibition on AI systems generating non-consensual intimate imagery and child sexual abuse material, effective December 2026. A legislature purely in deregulatory mode does not add prohibitions.

Where we part company with the optimistic reading is on the arithmetic. If the priority standards land in late 2026 and citation in the Official Journal follows, the practical runway to December 2027 is roughly twelve months of standards-backed preparation, not the sixteen the calendar suggests. Deferral bought time; it did not buy proportionally more usable time.

---

What This Means in Practice

For product and engineering

Treat disclosure as an architectural property, not a UI element. Marking applied at the render layer is lost the moment content leaves your surface. Marking applied at generation, propagated through your storage and delivery pipeline, and preserved by your CDN and CMS, is the version that survives. Audit whether your own infrastructure strips the manifests you are legally required to embed — in our reading, this is the most commonly overlooked Article 50 failure mode.

For compliance and legal

The high-risk deferral changes the deadline, not the requirement. Risk management, data governance, technical documentation, logging and human oversight survive intact; only the date moved. Inventory and classification work done for August 2026 retains its value. Note also that classification is not static: Article 111's transitional treatment of systems already on the market is conditioned on the absence of substantial modification, so an active roadmap can pull a legacy system back into scope.

For executives and boards

Ask one question of any AI governance report you receive this quarter: does this describe what our systems disclose, or what our systems decide? If the answer is entirely the former, the report is describing August 2026 compliance and nothing about December 2027 readiness. Those are different programmes with different cost structures.

For anyone consuming content

Calibrate on the asymmetry. A valid credential is meaningful evidence of origin. A missing credential is evidence of almost nothing.

Figure 2 — Diagram brief for the OneWise design team

Title: Where Disclosure Breaks: A Provenance Survival Map Purpose: Show at which pipeline stages a provenance signal is preserved or lost. Structure: Left-to-right pipeline with six stages: Generation → Internal storage → CMS / editorial tooling → CDN and transcoding → Third-party platform upload → Viewer screen. Dual signal rails beneath the pipeline: an upper rail for signed manifest and a lower rail for embedded watermark, each rendered as a continuous line that changes state at each stage. State encoding: Solid line = preserved; dashed = at risk; broken with a gap = typically lost. The manifest rail breaks at CDN and transcoding and again at platform upload; the watermark rail continues but thins to dashed after transcoding, with a break annotated "adaptive removal attack." Side branch: From Viewer screen, an arrow labelled "screenshot" loops back to a new node, "Re-captured file — both signals absent." Caption: Both mechanisms degrade, but they degrade at different points. That is the argument for layering — and the reason no layer should be read as proof of absence.

---

Key Insights

1. The first globally binding AI deadline regulated disclosure, not consequential decision-making.
2. Article 50 shipped on time because it needed guidance; Annex III slipped because it needed standards.
3. Harmonised standards, not political will, were the binding constraint — and Article 40's presumption of conformity is why they matter.
4. Transparency breaches carry the same headline penalty tier as high-risk breaches: €15m or 3% of turnover.
5. California and the EU converging on one date signals a de facto global baseline for generative-AI disclosure.
6. Provenance metadata proves a positive claim by a cooperating signer; it does not detect deception.
7. Peer-reviewed work shows invisible watermarks are removable under regeneration attacks — layering is a mitigation, not a fix.
8. Reading "unmarked" as "authentic" is the most damaging error a verification workflow can institutionalise.
9. The AI Omnibus added a prohibition as well as extending deadlines; it was a rebalancing, not a rollback.
10. Because standards arrive late, the deferral yields roughly twelve months of standards-backed runway, not sixteen.

---

Frequently Asked Questions

What changed on 2 August 2026? The EU AI Act's Article 50 transparency obligations became applicable, and the AI Office and national authorities began enforcement. California's AI Transparency Act became operative the same day.

Were the EU's high-risk AI rules delayed? Yes. Regulation (EU) 2026/1744 moved Annex III standalone high-risk obligations to 2 December 2027 and Annex I in-product obligations to 2 August 2028.

Was the whole AI Act delayed? No. Prohibited practices, AI literacy duties, GPAI model obligations, the penalty framework and Article 50 all remain in force on their original dates.

Does Article 50 apply to companies outside the EU? It can. The AI Act's scope follows the EU market. Organisations placing AI systems on that market or whose output is used there may be caught regardless of where they are established. Take specific advice.

What is the penalty for an Article 50 breach? Up to €15,000,000 or 3% of total worldwide annual turnover for the preceding financial year, whichever is higher, under Article 99(4)(g). For SMEs and start-ups the lower of the two figures applies.

Do I need to label AI-assisted text? Article 50 addresses text published to inform the public on matters of public interest, alongside synthetic image, audio and video. Light editing assistance on internal documents is a different case from publishing generated public-interest content. Scope determinations are fact-specific.

Is C2PA legally required? No. The law specifies outcomes — machine-readable marking that is effective, interoperable, robust and reliable as far as technically feasible — not a named specification. C2PA is the leading way to meet it, not the only permitted one.

If my chatbot is a third-party product, am I off the hook? Not automatically. Providers and deployers hold different duties. The organisation putting an assistant in front of users generally carries the disclosure obligation toward those users.

What is the AI Omnibus? Regulation (EU) 2026/1744, in force 27 July 2026. It amended the AI Act: it deferred high-risk application dates, added a prohibition on AI-generated non-consensual intimate imagery and CSAM effective December 2026, reinforced the AI Office's powers, extended simplified requirements to small mid-caps, and broadened regulatory sandbox access.

Why were the harmonised standards late? The scope is unusually broad, several requirements have no existing state of the art to codify, and the committee is among the largest and most diverse in European standardisation. CEN-CENELEC adopted exceptional acceleration measures in October 2025 targeting Q4 2026 availability for priority deliverables.

Does a missing Content Credential mean content is human-made? No. It may simply pre-date provenance systems, come from a non-participating tool, or have passed through a platform that strips metadata.

Can invisible watermarks be removed? Published peer-reviewed work demonstrates removal of pixel-level invisible watermarks through regeneration attacks. Assume a determined adversary can defeat a single mechanism.

Does California's law apply to small AI companies? Its covered-provider duties apply above a threshold of one million monthly visitors or users in California, so smaller providers generally fall outside — though later phases reach platforms, hosting services and device manufacturers.

Should we now slow down our high-risk compliance work? We would not. The substantive requirements were not softened, and the usable standards window is shorter than the calendar implies.

Is ISO/IEC 42001 sufficient for AI Act conformity? No. It is a credible AI management system framework and useful preparation, but only harmonised standards cited in the Official Journal confer the Article 40 presumption of conformity.

Where can I check current dates authoritatively? The Official Journal texts of Regulation (EU) 2024/1689 and Regulation (EU) 2026/1744, and the Commission's AI Act implementation pages.

---

Glossary

• AI Act — Regulation (EU) 2024/1689, the EU's horizontal AI framework, in force 1 August 2024.
• AI Office — The European Commission body supervising and enforcing AI Act rules, with specific powers over general-purpose AI models.
• AI Omnibus — Regulation (EU) 2026/1744, in force 27 July 2026, amending the AI Act's timelines and scope.
• Annex I / Annex III — The AI Act's two high-risk routes: systems embedded in regulated products, and standalone systems in listed sensitive areas.
• Article 40 presumption of conformity — The legal effect by which meeting a cited harmonised standard is presumed to satisfy the corresponding requirement.
• Article 50 — The AI Act's transparency provisions on AI interaction disclosure, synthetic content marking and deepfake labelling.
• C2PA / Content Credentials — An open specification for cryptographically signed provenance metadata describing origin and edit history.
• CAITA — The California AI Transparency Act, SB 942 as amended by AB 853.
• CEN-CENELEC JTC 21 — The joint technical committee developing European AI standards, including harmonised standards for the AI Act.
• Deployer — A party using an AI system under its own authority in a professional capacity.
• GPAI — General-purpose AI model, capable of a wide range of tasks and typically serving as a base for downstream systems.
• Harmonised standard — A European standard developed on a Commission request and cited in the Official Journal, carrying legal presumption effects.
• Latent disclosure — A machine-readable, imperceptible provenance signal embedded in content, as distinct from a visible label.
• Manifest disclosure — A visible, human-perceptible indication that content is AI-generated.
• Notified body — An accredited third party performing conformity assessments where the law requires independent review.
• Provider — The party developing an AI system or model and placing it on the market under its own name or trademark.
• Regeneration attack — A watermark-removal technique that noises and reconstructs content using a generative model.
• Soft binding — A durable in-content signal, such as a watermark or fingerprint, used to relink content to provenance data after metadata loss.
• Standardisation request (M/593, M/613) — The Commission mandates instructing CEN-CENELEC to develop AI Act standards.

---

Internal Linking Suggestions (OneWise)

• Category hub: Technology → AI Governance & Policy
• Explainer: What Is the EU AI Act? A Plain-Language Guide to Regulation (EU) 2024/1689
• Explainer: Content Provenance Explained: C2PA, Watermarking and What Each One Proves
• Guide: Building an AI System Inventory: A Practical Method for Risk Classification
• Guide: ISO/IEC 42001 and the AI Act: What an AI Management System Does and Does Not Cover
• Related commentary: Why Technical Standards Decide More AI Policy Than Legislatures Do
• Regional: US State AI Laws: A Tracking Reference for Compliance Teams

External Authority Link Suggestions

• European Commission, AI Act policy page — digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
• EUR-Lex, Regulation (EU) 2024/1689 (Official Journal text)
• EUR-Lex, Regulation (EU) 2026/1744 (AI Omnibus, Official Journal text)
• European Commission, AI Act Service Desk
• CEN-CENELEC, Artificial Intelligence / JTC 21 area of work
• California Legislative Information (leginfo), SB 942 and AB 853
• NeurIPS 2024 proceedings (watermark removal research)

Suggested "People Also Ask" Questions

• What changed in the EU AI Act on 2 August 2026?
• Have the EU high-risk AI rules been delayed?
• What are the fines for breaching AI Act Article 50?
• Do I have to label AI-generated images in the EU?
• What is the California AI Transparency Act?
• Is C2PA required by law?

Technical SEO Recommendations

• Schema markup: Article (or OpinionNewsArticle) with datePublished 2026-08-11, plus FAQPage on the FAQ block, BreadcrumbList, and DefinedTermSet on the glossary.
• Featured-snippet targets: the FAQ answers on the 2 August 2026 change, the delay question and the penalty question are each written as self-contained paragraph answers; Table 1 targets a table snippet for "EU AI Act timeline 2026".
• Secondary keywords: EU AI Act Article 50, AI Omnibus Regulation 2026/1744, California AI Transparency Act, AI content labelling requirements, high-risk AI obligations 2027.
• Long-tail keywords: what changed in the EU AI Act on 2 August 2026; are EU high-risk AI rules delayed to 2027; Article 50 fines €15 million 3% turnover; does a missing Content Credential mean content is real.
• Entities to reinforce: European Commission, AI Office, CEN-CENELEC JTC 21, C2PA, SynthID, ISO/IEC 42001, Official Journal of the European Union.
• Image alt text: (1) "Timeline diagram comparing EU AI Act disclosure obligations applicable August 2026 with high-risk obligations deferred to December 2027"; (2) "Pipeline diagram showing where C2PA provenance metadata and invisible watermarks are lost during content distribution"; (3) "Comparison table of EU AI Act Article 50 and the California AI Transparency Act".

---

References

Legislation and Official Documentation

• California Legislature. (2024). Senate Bill 942: California AI Transparency Act. Chapter 25 (commencing with Section 22757), Business and Professions Code. https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202320240SB942
• California Legislature. (2025). Assembly Bill 853: California AI Transparency Act. Chapter 674, Statutes of 2025. https://calmatters.digitaldemocracy.org/bills/ca_202520260ab853
• European Commission. (2026a). Commission starts enforcing AI Act rules and new transparency requirements on 2 August [Press release, 31 July 2026]. https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august
• European Commission. (2026b). AI Act [Policy page, last updated 3 August 2026]. Shaping Europe's Digital Future. https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
• European Commission. (2026c). Guidelines on transparency obligations for providers and deployers of certain AI systems [Published 20 July 2026]. https://digital-strategy.ec.europa.eu/en/policies/guidelines-transparency-ai-generated-content
• European Union. (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union, L 2024/1689. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=OJ:L_202401689
• European Union. (2026). Regulation (EU) 2026/1744 amending Regulation (EU) 2024/1689 (AI Omnibus). Official Journal of the European Union, L 2026/1744. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=OJ:L_202601744

Standards Bodies

• CEN-CENELEC. (2025, October 23). Update on CEN and CENELEC's decision to accelerate the development of standards for artificial intelligence. https://www.cencenelec.eu/news-events/news/2025/brief-news/2025-10-23-ai-standardization/
• CEN-CENELEC. (n.d.). Artificial Intelligence — CEN/CLC/JTC 21 area of work. Retrieved 11 August 2026, from https://www.cencenelec.eu/areas-of-work/cen-cenelec-topics/artificial-intelligence/

Academic and Conference Papers

• Christ, M., Gunn, S., & Zamir, O. (2024). Undetectable watermarks for language models. Proceedings of the 37th Annual Conference on Learning Theory (COLT 2024). https://arxiv.org/abs/2306.09194
• Kirchenbauer, J., Geiping, J., Wen, Y., Katz, J., Miers, I., & Goldstein, T. (2023). A watermark for large language models. Proceedings of the 40th International Conference on Machine Learning (ICML 2023). https://arxiv.org/abs/2301.10226
• Saberi, M., Sadasivan, V. S., Rezaei, K., Kumar, A., Chegini, A., Wang, W., & Feizi, S. (2024). Robustness of AI-image detectors: Fundamental limits and practical attacks. International Conference on Learning Representations (ICLR 2024). https://arxiv.org/abs/2310.00076
• Zhao, X., Zhang, K., Su, Z., Vasan, S., Grishchenko, I., Kruegel, C., Vigna, G., Wang, Y.-X., & Li, L. (2024). Invisible image watermarks are provably removable using generative AI. Advances in Neural Information Processing Systems 37 (NeurIPS 2024). https://proceedings.neurips.cc/paper_files/paper/2024/hash/10272bfd0371ef960ec557ed6c866058-Abstract-Conference.html

Scholarly Commentary and Analysis

• Cantero Gamito, M. (2025, November 28). From consensus to exceptionality: What the EU's AI standards crisis reveals about delegated technical governance. RealAw Blog. https://realaw.blog/2025/11/28/from-consensus-to-exceptionality-what-the-eus-ai-standards-crisis-reveals-about-delegated-technical-governance-by-marta-cantero-gamito/
• Morgan Lewis. (2026, August). New California AI disclosure rules become operative. https://www.morganlewis.com/pubs/2026/08/new-california-ai-disclosure-rules-become-operative

Note on Sources

All dates and legal thresholds in this article were verified against primary legislative texts or official European Commission and Californian legislative publications as at 11 August 2026. Where analysis rests on secondary legal commentary, it is attributed as such. Regulatory timelines in this area have changed more than once; readers making compliance decisions should confirm current dates against the Official Journal and take qualified advice on their own circumstances.