17 January 2028 sounds a long way off. For any EU financial entity designated for DORA TLPT (Threat-Led Penetration Testing), it isn't. Once you account for provider scarcity, regulatory scoping, and a testing cycle that runs 9 to 14 months on its own, the real deadline that matters is 2026 — because that is when procurement has to begin.

If your bank, insurer, investment firm, or payment institution has received a designation notice from your National Competent Authority (NCA), this article breaks down exactly what Threat-Led Penetration Testing under DORA Article 26 requires, why the timeline is tighter than it looks, and what to do about it right now.

What Is DORA TLPT, Exactly?

The Digital Operational Resilience Act (Regulation (EU) 2022/2554) has been in force across the EU since 17 January 2025. Among its five pillars — ICT risk management, incident reporting, resilience testing, third-party risk, and information sharing — Article 26 and Article 27 introduce the most demanding obligation of all: Threat-Led Penetration Testing, modelled directly on the European Central Bank's TIBER-EU framework.

Unlike a standard vulnerability scan or annual penetration test, TLPT is an intelligence-led, covert red team exercise run against your live production environment. Your own security operations team is not told it is happening. A licensed threat intelligence provider first builds a Targeted Threat Intelligence (TTI) report profiling the real adversaries most likely to target your institution — nation-state actors, organised financial cybercrime groups, or insider-threat scenarios. An accredited red team then executes those exact attack scenarios against your critical or important functions, including outsourced and cloud infrastructure, for a minimum of 12 weeks.

TLPT is not optional and it is not self-selected. Your NCA designates you based on systemic importance, asset size, and criticality to the financial system. Once designated, the obligation repeats at least every three years, and no generic penetration test can substitute for it.

The 2028 Deadline — And Why It's Already Close

The first mandatory TLPT cycle under DORA must be completed by 17 January 2028. On paper, that is more than a year away from today. In practice, a full engagement — provider procurement, scope agreement with your competent authority, the threat intelligence phase, the red team campaign, purple teaming, remediation, and final attestation — typically takes between 9 and 14 months once everything is running smoothly.

Typical DORA TLPT Timeline

Work backwards from January 2028 and the math is unforgiving: procurement should realistically start in 2026, not 2027. Entities that wait until designation pressure builds risk being left with whatever accredited provider capacity remains — often at a premium, and often without the specialist industry experience their scope actually needs.

What Happens If You Miss the Deadline?

Missing the 2028 deadline, running a poorly scoped test, or failing to remediate critical findings on the agreed timeline all expose an institution to enforcement action under DORA Article 50, including financial penalties tied to global annual turnover and operational restrictions imposed by supervisors. For designated entities, TLPT sits alongside the broader resilience testing programme required under Article 25 — but it carries a legal weight and reputational visibility that an annual vulnerability scan does not.

How to Prepare Now

• Confirm designation status with your NCA and don't assume you're out of scope simply because you haven't been formally notified yet.
• Start provider procurement in 2026 — evaluate TIBER-EU accredited threat intelligence and red team providers before capacity dries up.
• Map critical and important functions, including third-party and cloud dependencies, ahead of scoping discussions.
• Run standing red team and continuous penetration testing programmes so TLPT becomes a checkpoint rather than a scramble.
• Align TLPT with your wider DORA programme — ICT risk management, incident reporting, and third-party risk registers all feed into a credible scope document.

This is exactly where experienced penetration testing services earn their keep well before the formal TLPT clock starts. A mature, continuous testing programme built on CREST-approved methodology gives your institution a defensible baseline while you queue for accredited TLPT capacity. VistaInfoSec's broader guidance on common DORA compliance challenges is a useful starting point if you're still building out your resilience testing roadmap.

It's also worth understanding how TLPT fits alongside your other frameworks. If your institution already holds ISO 27001 or SOC 2, this DORA, ISO 27001 and SOC 2 mapping guide shows exactly where DORA's testing requirements go beyond what those certifications already cover. And if NIS2 obligations apply to any part of your group alongside DORA, this NIS2 vs DORA compliance guide untangles where the two regulations overlap and where they diverge.

The Bottom Line

DORA TLPT isn't a 2028 problem — it's a 2026 decision. The financial entities that treat threat-led penetration testing as a checkpoint within an already-mature security testing programme will move through designation, scoping, and attestation calmly. Those that wait will be negotiating with whatever accredited provider has a slot left, on someone else's timeline. For EU financial institutions serious about operational resilience, the smartest move this year is simple: start the conversation with accredited providers now, not in Q4 2027.

Three months of "not yet live" updates end this week. ENISA has scheduled the Single Reporting Platform (SRP) — the mandatory channel for Cyber Resilience Act incident and vulnerability notifications — to go operational on 11 September 2026, the exact date Article 14 reporting duties enter into application. For European developers, product security teams, and manufacturers placing connected products on the EU market, this is no longer a compliance deadline on a roadmap slide. It is a live, 24-hour regulatory clock.

If your team builds IoT devices, embedded firmware, industrial controllers, or software with a network connection sold anywhere in the EU, this article breaks down exactly what the Cyber Resilience Act (Regulation (EU) 2024/2847) requires you to report, when, and through which channel — without the guesswork.

What Is the ENISA Single Reporting Platform?

The Single Reporting Platform is the electronic entry point established under Article 16 of the CRA. Instead of notifying multiple national authorities separately, a manufacturer files one report through the SRP, which then routes it simultaneously to the national CSIRT designated as coordinator for the manufacturer's main EU establishment, and to ENISA itself.

Crucially, ENISA has confirmed that no reporting API will be offered at this stage. Submissions go through a browser-based interface, authenticated via EU Login, the European Commission's shared sign-in credential. Internal detection and triage workflows can be automated, but the final act of filing is manual. Teams that assumed system-to-system integration would carry them through launch week need to rebuild that assumption immediately.

What Developers Must Actually Report Under Article 14

Article 14 does not ask you to report every bug ticket. It creates two distinct reporting tracks, and understanding the difference is the single most important thing a developer or product security lead can do before touching the platform.

1. Actively exploited vulnerabilities. If a vulnerability in your product with digital elements is being exploited in the wild — not theoretical, not internally discovered during a code review, but actually under active exploitation — you are obligated to notify ENISA and your coordinating CSIRT.

2. Severe incidents having an impact on the security of the product. This covers incidents that materially affect the product's ability to protect the confidentiality, integrity, or availability of data or functions.

For both tracks, the CRA imposes a rigid three-stage timeline:

• 24 hours — an early warning, submitted as soon as the manufacturer becomes aware.
• 72 hours — a more detailed notification confirming severity, indicators of compromise, and corrective measures underway.
• 14 days — a final report once the vulnerability or incident is resolved, including a root-cause assessment.

This is precisely why compliance specialists have been urging manufacturers to treat the reporting obligation as a distinct, earlier milestone from the CRA's broader December 2027 conformity deadline. VISTA InfoSec's detailed Cyber Resilience Act compliance checklist walks through exactly how to sequence this preparation across scoping, documentation, and reporting readiness.

Who Is on the Hook — and Who Isn't

Manufacturers carry almost the entire reporting burden. Importers and distributors have separate, narrower duties around conformity verification and CE marking, but the 24/72-hour reporting clock sits squarely with the manufacturer of the product with digital elements. Open-source software stewards also fall within ENISA's guidance and should register and validate on the SRP only when they have an actual notification to file, to avoid overloading national CSIRT teams during the launch window.

For manufacturers not established in the EU, the coordinating CSIRT is determined by the country of the manufacturer's authorised representative — not by where the company is headquartered. This detail catches non-EU vendors off guard more often than any other part of Article 14.

What Developers Should Prepare Before Filing a First Report

Even with the platform now operational, several practical gaps remain open. The official reporting data format is set by a separate European Commission implementing act, and full registration handbooks, dry-run environments, and the finalised list of national CSIRT coordinators have been rolling out only in the weeks immediately before launch.

What development and product security teams can control right now:

• Create EU Login accounts in advance for a primary and secondary organisational representative, using durable work email addresses.
• Maintain a live product inventory mapping every shipped SKU to a manufacturer identity, an SBOM, and a named PSIRT contact.
• Build (and rehearse) the internal detection-to-decision workflow that determines, within hours, whether an event meets the "actively exploited" or "severe incident" threshold.
• Name a reporting owner and a documented backup, since the 24-hour clock does not pause for someone being on leave.
• Run a full dry-run submission the moment ENISA's test environment is accessible, rather than waiting for a real incident to be your team's first encounter with the interface.

Because classification (Default, Important I, Important II, or Critical under Annex III/IV) determines both your conformity route and how deep your documentation needs to go, organisations still mapping their product portfolio against these categories should not wait. VISTA InfoSec's Cyber Resilience Act gap assessment guide is a useful reference for structuring that inventory and documentation work alongside reporting readiness, rather than treating them as separate projects.

Penalties Make This Non-Negotiable

Non-compliance with the CRA carries fines of up to €15 million or 2.5% of global annual turnover, whichever is higher — figures that place it firmly alongside GDPR and NIS2 in terms of enforcement weight. For a regulation whose full conformity requirements do not bite until December 2027, it is notable that the reporting obligation — arguably the operationally hardest part to get right under time pressure — is the piece that started first.

The Bottom Line for European Developers

The SRP going live does not mean your compliance work is finished; it means the clock that actually matters has started. Article 14 rewards organisations that have already separated "actively exploited vulnerability" from routine bug triage, pre-built their EU Login access, and rehearsed their escalation path. Teams still treating CRA reporting as a 2027 problem are now operating under a live regulatory deadline with real financial exposure.

Europe's automotive supply chain has spent a decade tightening its grip on data security, and the next milestone has a hard date attached to it. Mercedes-Benz has confirmed that its dealer and supplier network must demonstrate a certified information security programme either ISO 27001 or TISAX Level 2 by September 30, 2026. For anyone connected to the Mercedes-Benz ecosystem, this is no longer a "nice to have." It is a contractual condition of staying in business with one of the world's most recognisable car makers.

If that sentence made your compliance team sit up a little straighter, good. It should. Let's unpack exactly what is changing, why it matters so much to European suppliers and dealers, and how to get certification-ready before the clock runs out.

Why Mercedes-Benz Is Tightening the Screws on Cybersecurity

The automotive industry has learned some hard lessons about supply chain risk. The 2024 CDK Global ransomware incident, which knocked more than 15,000 dealerships offline across North America, is the case study every OEM security team now references. Attackers rarely go straight for a manufacturer's own network they look for the weakest link, often a smaller partner with looser controls, and use that as a launchpad into the parent company's systems.

Mercedes-Benz's response mirrors what German OEMs have been doing for years through TISAX (Trusted Information Security Assessment Exchange), the automotive industry's shared assessment framework built by the VDA (German Association of the Automotive Industry) and operated by the ENX Association. TISAX already underpins security expectations across Volkswagen Group, BMW, Audi, Porsche and their extended supplier base, and Mercedes-Benz is now applying that same logic verified, independent proof of security — to its own network rather than accepting self-attestations.

What the September 30, 2026 Requirement Actually Says

Mercedes-Benz is not mandating a single rigid path. Organisations in scope can satisfy the requirement in one of two recognised ways:

• ISO/IEC 27001 certification — the internationally recognised standard for building and operating an Information Security Management System (ISMS), applicable across any industry.
• TISAX Assessment Level 2 (AL2) — the automotive-specific assessment run through the ENX portal, built on the VDA-ISA control catalogue, which itself draws heavily on ISO 27001/27002 principles with automotive-specific additions such as prototype protection and connected-vehicle data handling.

Either path counts as evidence of a "qualified information security programme." What no longer counts is a checklist, a vendor questionnaire filled out by an internal team, or software that claims compliance without an independent audit trail. The deadline applies at an organisational level, and Mercedes-Benz — like Stellantis, which has set an identical September 30, 2026 deadline for its own supplier base — expects the certificate or TISAX label to be in hand, not "in progress," by that date.

Why This Matters More for European Suppliers Than It Might Seem

It's tempting to read "Mercedes-Benz dealer network" and assume this is a North American story. It isn't, not really. TISAX itself is a distinctly European mechanism, born in Germany and already deeply embedded in the operations of Mercedes-Benz, BMW, Volkswagen, Audi and Porsche's European supply chains. What is happening now is the same discipline being extended further down the chain and applied with a hard, enforced deadline rather than a soft recommendation.

For European Tier 1 and Tier 2 suppliers, marketing agencies handling prototype imagery, logistics partners, and IT service providers touching Mercedes-Benz data anywhere in the value chain, this is a signal worth reading closely: the era of "we'll get to it eventually" is over. Contracts are increasingly being written with certification as a condition precedent, not a follow-up item.

ISO 27001 or TISAX — Which Should You Choose?

This is the question every compliance lead is currently wrestling with, and the honest answer is: it depends on who you sell to.

• If your relationships extend beyond the automotive sector — to finance, healthcare, SaaS customers, or public sector contracts — ISO 27001 gives you a globally recognised certificate that opens doors well beyond Mercedes-Benz.
• If your business is automotive-specific and you already work with, or hope to work with, multiple German OEMs, TISAX lets you complete one assessment and share the resulting label across Mercedes-Benz, BMW, VW Group and others through the ENX portal — avoiding repeated audits for each relationship.
• Many organisations that already hold ISO 27001 find that TISAX readiness moves noticeably faster, since the risk assessment methodology, policies and core Annex A controls are already built and operating.

Timelines matter here too. Starting from scratch, most organisations need anywhere from four to twelve months to reach a TISAX label or ISO 27001 certificate, with the assessment itself typically booked weeks in advance. With September 30, 2026 on the calendar, the realistic window to start a programme from zero and still land the certification comfortably before the deadline is closing fast.

Getting Certification-Ready Without the Guesswork

The path to either certification generally follows the same shape: a gap assessment against the relevant control catalogue (ISO 27001 Annex A or VDA-ISA), remediation of the gaps that surface, implementation of documented policies and evidence trails, and finally the formal audit through an accredited certification body or an ENX-accredited TISAX audit provider.

Organisations that try to run this entirely in-house often underestimate how much evidence collection and internal alignment it takes to pass a Stage 1/Stage 2 ISO 27001 audit, or a TISAX AL2 assessment, on the first attempt. That is exactly the gap that specialist advisory firms exist to close. VISTA InfoSec's ISO 27001 Advisory & Certification service works alongside internal teams to design the ISMS, run the risk assessment, and prepare for Stage 1 and Stage 2 audits without forcing a generic template onto your business. For organisations that sell specifically into the German and European automotive supply chain, VISTA InfoSec's TISAX Audit & Certification practice in Germany runs VDA-ISA gap assessments, scopes the correct assessment level, and manages ENX portal registration end to end.

If you're still weighing which certification actually fits your business model, this detailed breakdown of TISAX vs ISO 27001 for automotive suppliers is a useful next read it compares governing bodies, scope, cost drivers and typical timelines side by side.

The Bottom Line

September 30, 2026 is not a soft target it's a contractual deadline set by one of the automotive world's most demanding customers, echoed almost identically by Stellantis. Whether your organisation ultimately pursues ISO 27001 or TISAX Level 2, the underlying message from Mercedes-Benz is the same one German OEMs have been sending their supply chains for years: prove it, don't just promise it. Suppliers and dealers who start their gap assessment now will spend 2026 building a defensible security programme. Those who wait may find themselves racing an audit calendar that has already filled up.

Somewhere inside your organisation, an employee has connected a generative AI tool to a customer database. A marketing assistant has plugged an autonomous agent into your CRM to "save time." A developer has wired an MCP server into your production pipeline over a weekend sprint. Nobody filed a request. Nobody ran a risk assessment. Nobody in security even knows it happened.

This is shadow AI and in 2026, it is no longer a fringe IT hygiene issue. It is the single fastest-growing compliance exposure for European businesses, and regulators are catching up faster than most boardrooms realise.

What Exactly Is a Shadow AI Agent?

Shadow AI refers to AI tools, models, or autonomous agents used inside a business without the knowledge, approval, or oversight of IT and security teams. It has evolved well beyond an employee pasting text into a public chatbot. Today's shadow AI increasingly means agentic AI autonomous software that can log into systems, call APIs, move data between platforms, and take actions with little or no human review, often via the Model Context Protocol (MCP).

Gartner projects that by the end of 2026, 40% of enterprise applications will feature task-specific AI agents, up from under 5% in 2025 and a significant share of those deployments will happen outside any formal security review. Traditional shadow IT exposed unapproved software. Shadow AI agents expose live data pipelines, credentials, and decision-making authority to systems nobody has vetted.

Why This Has Become a European Board-Level Problem

For companies operating in or serving the EU, this isn't an abstract cyber-risk conversation anymore — it's a regulatory one. Two frameworks now converge on the same blind spot:

The GDPR angle. Any shadow AI tool that processes customer, employee, or prospect data is a personal data processing activity, whether or not it was ever declared. If that tool retains prompts, trains on inputs, or transfers data outside the EU, it can trigger GDPR obligations around lawful basis, data minimisation, and cross-border transfer with fines of up to €20 million or 4% of global annual turnover for serious breaches.

The EU AI Act angle. The AI Act (Regulation (EU) 2024/1689) is now live in phases. Prohibited-practice rules have been enforceable since February 2025, and obligations for general-purpose AI providers took effect in August 2025. Following the Digital Omnibus on AI finalised in the Official Journal in July 2026 the compliance deadline for most high-risk Annex III systems has moved to December 2027, but the Article 50 transparency obligations covering chatbots and AI-generated content remain due from August 2026. Crucially, prohibited-practice penalties already run as high as €35 million or 7% of global turnover, a ceiling that exceeds even GDPR's. An unregistered, ungoverned agent quietly making HR, credit, or profiling decisions could already sit in a high-risk category regulators are actively watching.

The regulatory direction is unambiguous: the EU AI Act does not replace GDPR, it sits alongside it. A shadow agent that violates one is very likely violating both.

The Scale of the Problem Is Bigger Than Most CISOs Think

Recent industry research paints a sobering picture for 2026:

• Shadow AI incidents are projected to triple by the end of 2026, according to Gartner, with an estimated 25–35% of enterprise AI spend occurring entirely outside IT visibility.
• MCP-based agent adoption grew more than 400% in 2025, with the majority of deployments occurring outside any formal security review.
• Only one in five organisations reports having a mature governance model for autonomous AI agents, according to Deloitte's 2026 State of AI in the Enterprise report.
• Roughly 98% of organisations report some form of unsanctioned AI use, and nearly half expect a shadow AI-related incident within the next twelve months.

For a European business, every one of these agents is also a potential GDPR processing activity and a potential AI Act touchpoint that has never been assessed, documented, or registered.

How to Find Shadow AI Agents Before a Regulator Does

The organisations getting ahead of this are treating shadow AI discovery the same way they'd treat any other compliance audit structured, evidence-based, and continuous.

1. Run a full AI and data-flow inventory. You cannot govern what you cannot see. Map every AI tool, browser extension, API key, and MCP server connected to company systems including tools bundled inside vendor software you already use.
2. Classify by data sensitivity and decision authority. Not every AI tool carries the same risk. Prioritise agents that touch personal data, financial data, or make autonomous decisions affecting individuals these are the ones GDPR and the EU AI Act care about most.
3. Conduct a Data Protection Impact Assessment (DPIA) wherever personal data is involved. Under GDPR Article 35, any processing likely to result in high risk to individuals' rights requires a DPIA before not after deployment. Retrofitting one after discovery is still far better than having none at all.
4. Map agents against AI Act risk tiers. Determine whether any shadow agent could be classified as high-risk under Annex III (employment decisions, credit scoring, biometric processing) and document your reasoning either way regulators will ask for it.
5. Close the gap with policy, not prohibition. Outright bans consistently fail; usage simply moves to personal devices and becomes even less visible. Provide sanctioned, governed alternatives instead.
6. Build continuous monitoring, not a one-off sweep. Shadow AI reappears within weeks of any single audit unless detection is ongoing and tied into your existing security and privacy programme.

Turning Discovery Into Compliance

Finding shadow AI agents is only half the job. The other half is proving to a regulator convincingly and with documentation that you found them, assessed them, and controlled the risk. That means pairing technical discovery with a proper GDPR risk assessment, running a documented GDPR compliance audit, and understanding exactly when a DPIA is legally required under Article 35.

On the AI Act side, working through a structured EU AI Act compliance checklist helps European businesses classify agents correctly and avoid both prohibited-practice exposure and unnecessary over-compliance. If your organisation is weighing internal resourcing against external expertise, it's also worth reviewing what a realistic GDPR compliance budget looks like in 2026, since AI-specific impact assessments now add a meaningful line item to most privacy programmes.

The Bottom Line

Shadow AI agents are not a future risk for European companies they are running in production right now, often with more autonomy and system access than the shadow IT of a decade ago ever had. Regulators enforcing GDPR and the EU AI Act are not waiting for companies to volunteer this information; supervisory authorities are increasingly proactive, and the penalties on both sides of this overlap now rank among the highest in global regulation.

The organisations that will avoid the next headline fine are the ones auditing their AI footprint today not the ones waiting to be asked. Find the agents. Document the risk. Close the gap. Do it before your regulator does it for you.

If you work in AI governance anywhere near Europe, you have heard the sigh of relief that followed the EU's Digital Omnibus agreement in May 2026. The European Parliament's final approval in June 2026 pushed the compliance deadline for standalone high-risk AI systems under Annex III from 2 August 2026 to 2 December 2027 a sixteen-month reprieve. Annex I high-risk systems, embedded in regulated products like medical devices and machinery, now have until 2 August 2028.

For many compliance teams, that news landed like a permission slip to slow down. It shouldn't. The delay changes the calendar, not the destination and organisations that keep building their AI governance programme now, anchored around ISO 42001 certification, will be the ones still standing when enforcement actually begins.

What Actually Got Delayed — and What Didn't

It's worth being precise here, because the Digital Omnibus was not a blanket pause on the EU AI Act. Three things happened:

1. Annex III (use-based) high-risk obligations — covering employment, credit scoring, education, law enforcement, and critical infrastructure — moved from August 2026 to December 2027.
2. Annex I (product-embedded) high-risk obligations — radio equipment, lifts, medical devices — moved from August 2027 to August 2028.
3. National regulatory sandboxes that member states must operate were pushed back by a year, to August 2027.

What did not move: transparency obligations under Article 50, covering AI chatbots, deepfakes, and synthetic content labelling, remain enforceable, with the watermarking deadline actually tightened to 2 December 2026. Prohibited AI practices social scoring, manipulative systems, and the new ban on AI-generated non-consensual intimate imagery have applied since February 2025 and are not affected. General-purpose AI model obligations under the Act have applied since August 2025.

So the "delay" is really a targeted, staggered postponement of the hardest part: high-risk system conformity assessments. The reason is instructive too European standards bodies like CEN-CENELEC simply haven't finished the harmonised technical standards that high-risk providers need to demonstrate conformity against. The law didn't get easier; the infrastructure to comply with it wasn't ready.

Why ISO 42001 Is the Bridge Between the Two

This is exactly where ISO/IEC 42001:2023, the world's first certifiable standard for an Artificial Intelligence Management System (AIMS), earns its keep. It gives organisations a structured, auditable framework covering AI risk assessment, data governance, human oversight, transparency, and lifecycle monitoring that maps closely onto the same obligations the EU AI Act eventually requires for high-risk systems.

Put simply: ISO 42001 doesn't replace the AI Act, and certification alone won't satisfy every legal obligation. But it is the most efficient way to build the actual governance muscle documented risk management, impact assessments, monitoring, and accountability that regulators, auditors, and enterprise customers will expect to see regardless of which deadline applies to you. A well-run ISO 42001 certification process, typically taking four to twelve months from gap assessment to certificate, builds exactly the documentation trail that Annex III providers will need in 2027 anyway.

Three Reasons Certifying Now Still Makes Sense

1. The delay is a runway, not a reprieve. Sixteen months sounds generous until you map it against a realistic certification timeline. Between the Stage 1 and Stage 2 audits, gap remediation, and the technical standards still catching up, organisations that wait until late 2027 to start are gambling against a hard deadline with no further extensions expected.

2. Procurement doesn't wait for regulators. Enterprise buyers across Europe are already asking vendors for AI governance evidence before signing contracts regardless of what the statutory deadline says. An ISO 42001-certified AI management system replaces a hundred repetitive security questionnaires with one recognised certificate, which is a commercial advantage today, not in 2027.

3. Fines for prohibited practices and transparency failures are live now. Penalties for banned AI practices reach €35 million or 7% of global turnover, and transparency violations are enforceable immediately. A functioning AIMS built around ISO 42001 principles gives you the risk register, oversight structure, and audit trail to catch these issues before a regulator does not just the eventual high-risk classification.

Building a Governance Programme That Covers Both

The smartest path for European AI providers and deployers right now isn't choosing between ISO 42001 and EU AI Act compliance it's building one programme that satisfies both. A practical way to structure this:

• Start with an AI system inventory and risk classification, exactly as recommended in a thorough EU AI Act compliance checklist, so you know today which systems are high-risk under Annex III and which face the earlier, unaffected obligations.
• Build the management system policies, risk treatment, human oversight, and monitoring against the ISO 42001 clauses, since these controls map directly onto what Annex III will eventually demand.
• Run a gap assessment against the ten priority controls for EU AI Act readiness to close the distance between where your AIMS is today and where the regulation will expect it to be.
• Treat certification as continuous, not a one-time event annual surveillance audits under ISO 42001 keep the system current as the EU AI Office issues further guidance through 2026 and 2027.

The European Angle: Trust Is the Real Currency

For organisations operating across the EU, the calculus isn't only regulatory. European customers, works councils, and public-sector procurement teams are increasingly wary of AI systems they can't audit. A certified AIMS signals something a compliance memo cannot: that your organisation treats AI risk as seriously as it treats financial risk or data protection disciplines Europe has already forced the world to take seriously through GDPR and NIS2.

The Digital Omnibus bought the market time to get the technical standards right. It did not buy providers a reason to stop building trust. Organisations that treat this window as free time will spend 2027 scrambling; those that treat it as a head start will spend 2027 renewing a certificate they already earned.

Final Word

The EU AI Act delay is real, and it's a sensible response to genuine standards-readiness problems in Brussels not a signal that AI governance can wait. If anything, it's the best argument yet for starting ISO 42001 certification now: you get a working AI management system, a competitive edge in EU procurement, and a running start on obligations that are still coming, just later than originally planned.

Brussels has finally put teeth into AI safety. If your organisation builds, deploys, or even evaluates general-purpose AI (GPAI) models with systemic risk, Article 55 of the EU AI Act is no longer a footnote in a compliance deck it is an operational requirement with a live enforcement clock. And as of August 2026, the European Commission's AI Office has the power to check whether you actually did the work.

For a continent that has spent two years debating what "trustworthy AI" means in practice, this is the moment theory turns into audit trails.

What Article 55 Actually Requires

Article 55 applies to a narrow but consequential group: providers of GPAI models classified as carrying systemic risk, typically because they cross the compute threshold set out in Article 51 or are formally designated by the Commission. Think frontier-scale foundation models from the handful of labs capable of training at that scale not the thousands of SMEs building applications on top of them.

For this tier, the obligation is unambiguous: providers must evaluate their models using state-of-the-art protocols, including adversarial (red-teaming) testing, to identify and mitigate systemic risks before those risks reach the Union market. That testing must be proportionate to the model's risk profile, may involve independent external experts, and has to cover misuse scenarios, dangerous capability evaluations, and vulnerability assessments then be documented and, where relevant, reported to the AI Office.

Alongside testing, Article 55 also obliges providers to assess and mitigate systemic risk at Union level, maintain adequate cybersecurity for the model and its infrastructure, and report serious incidents without undue delay. Mitigations can range from changing model architecture and adding safety mechanisms to restricting deployment altogether.

The Timeline Europe Actually Needs to Know

• 2 August 2025 — Article 55 obligations became legally applicable to systemic-risk GPAI providers.
• 2 August 2026 — The AI Office's enforcement powers kick in: formal information requests, mandated mitigation measures, and administrative fines.
• 2 August 2027 — Transitional deadline for GPAI models already on the market before August 2025.

The gap between 2025 and 2026 was never a grace period to relax — it was the runway for the AI Office to build supervisory capacity and for the GPAI Code of Practice's Safety and Security chapter to become the de facto rulebook. The May 2026 Digital Omnibus reinforced the AI Office's central supervisory role without pushing these dates back. If anything, Brussels tightened the loop.

What Adversarial Testing Actually Looks Like in Practice

This is where the regulation stops being abstract. Under Article 55, "adversarial testing" is not a single scan or a checkbox exercise — it is a structured, multi-layered discipline that mirrors mature cybersecurity red teaming far more than it resembles traditional software QA.

If that last point sounds familiar, it should. It is the same principle that has underpinned mature information security programmes for years: an internal team can harden a system, but only an independent, CREST-accredited red team can genuinely tell you where it breaks. Organisations that have already engaged professional red team assessment services for their IT infrastructure have a real head start, because the discipline of planning, reconnaissance, staged attack simulation, and documented findings translates directly into what GPAI providers now need to demonstrate under Article 55.

Why This Matters Beyond the Frontier Labs

Most European organisations are not training 10²⁵-FLOP models, so Article 55 will not apply to them directly. But the ripple effect is real. Deployers building on top of systemic-risk GPAI models will increasingly be asked by enterprise customers and auditors to show due diligence on the models they integrate including whether the underlying provider's Article 55 testing and Code of Practice commitments are credible. It's worth understanding how red team assessments differ from standard penetration testing, since the two are frequently confused in vendor questionnaires and procurement checklists.

The Bottom Line for 2026

Article 55 marks the point where the EU AI Act stopped being a documentation exercise and became a testing mandate with real enforcement muscle behind it. For the handful of frontier GPAI providers, adversarial testing must now be systematic, independently verifiable, and tied to concrete mitigation not a marketing claim in a model card. For everyone else in the European AI supply chain, the message is just as clear: red teaming is no longer optional cybersecurity best practice. It is fast becoming the shared language of AI accountability across the Union.

Organisations preparing for this shift whether validating a systemic-risk GPAI model or the infrastructure it runs on should treat independent adversarial testing as a standing programme, not a one-time audit.

For twelve months, Brussels asked nicely. As of August 2, 2026, it doesn't have to anymore.

If your compliance team spent the summer congratulating itself on a "quiet" AI Act rollout, it's time for an uncomfortable conversation. The obligations for general-purpose AI (GPAI) providers didn't just appear this month — they've technically applied since August 2, 2025. What changed on August 2, 2026 is that the European Commission's AI Office can finally do something about non-compliance: audit models, demand corrective action, restrict market access, and issue fines of up to €15 million or 3% of global annual turnover, whichever is higher.

That distinction — obligation versus enforcement — is exactly what most European boardrooms missed while they were busy tracking the wrong deadline.

The Grace Period Is Over

When the EU AI Act entered into force in August 2024, it built in a deliberate one-year runway for GPAI providers. Chapter V obligations — training-data summaries, copyright compliance, technical documentation, and systemic-risk management for the most powerful models — became legally binding on August 2, 2025. But the AI Office needed time to build its own supervisory machinery before it could act on any of it.

That runway has now ended. From August 2, 2026, the Commission can request technical documentation, run model evaluations, order risk-mitigation measures, and pull non-compliant GPAI models from the EU market. Models placed on the market before August 2, 2025 get a slightly longer runway — they must be fully compliant by August 2, 2027 — but every model launched after that date has already been operating on borrowed time.

Signing the voluntary GPAI Code of Practice helps, but it isn't a shield. Regulators have indicated that Code signatories will have their good-faith commitments weighed when calculating penalties, yet enforcement applies "signatures or not." A partial signature — committing to some chapters of the Code while skipping others — is a detail worth scrutinising in any vendor's compliance claims, not taking at face value.

Article 50 Is the Deadline Everyone Underestimated

While GPAI enforcement grabbed the headlines, Article 50 transparency obligations quietly went live the same day — and this one reaches far beyond model providers. Any organisation deploying a chatbot or conversational AI system must now disclose, clearly and at the start of the interaction, that the user is talking to an AI. AI-generated or manipulated content, including deepfakes, requires machine-readable labelling. The same €15 million / 3% turnover penalty ceiling applies here too.

This is the requirement that catches European businesses off guard, because it isn't aimed at Silicon Valley labs — it's aimed at every customer service bot, marketing assistant, and generative content workflow running inside ordinary companies across Germany, France, Ireland, and the Netherlands. If your team's AI inventory doesn't already flag which systems talk directly to customers, that's the gap to close first.

Don't Confuse This With the Digital Omnibus Delay

Here's where a lot of internal risk registers went wrong this year. The Digital Omnibus, finalised by the European Parliament on June 16, 2026 and given final Council sign-off on June 29, 2026, pushed high-risk AI system obligations under Annex III from August 2026 out to December 2, 2027. That's a genuine, significant delay — but it applies to a completely different track of the Act.

It does nothing to soften GPAI enforcement powers or Article 50 disclosure duties. If your compliance roadmap assumed the Omnibus bought extra breathing room on chatbot transparency, it was working from an outdated script. Two separate clocks, two separate consequences — and conflating them is precisely how well-resourced teams end up caught flat-footed on the deadline that actually mattered.

What Your Team Likely Missed

1. Treating "GPAI obligations" and "GPAI enforcement" as the same milestone. They weren't. The rules existed for a year with no teeth; now they have teeth.
2. Assuming the Omnibus delay covered everything. It covered high-risk systems only — not GPAI supervision, not Article 50.
3. Ignoring deployer-side exposure. Being a "deployer" rather than a "provider" doesn't create a safe harbour under Article 50. Disclosure duties reach anyone whose customers interact with AI.
4. No live AI inventory. Regulators, national market surveillance authorities, and even downstream providers can now trigger scrutiny. Without a current inventory of AI systems, owners, and purposes, an inquiry response starts from zero.
5. Reading "Code of Practice signatory" as full compliance. It's a mitigating factor in penalty calculation, not an exemption.

What To Do Before the Next Inquiry Lands

European organisations — not just AI labs — now sit inside an active enforcement regime. Practical next steps look less like a policy rewrite and more like an operational audit: build (or refresh) a complete AI systems inventory, classify which tools are GPAI-adjacent versus deployer-only, confirm chatbot disclosure is actually implemented at the interaction level, and stress-test whether your documentation would survive a Commission request this quarter.

The Bottom Line

August 2, 2026 didn't introduce new rules. It introduced consequences. For European businesses running customer-facing AI, procuring GPAI models, or quietly letting departments adopt generative tools without central oversight, the honest question isn't "are we compliant on paper" — it's "could we produce evidence of compliance inside a week if the AI Office asked." If the answer is uncertain, the runway to find out just got a great deal shorter.