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.