The Deadlines Are Now
For AWS customers in Europe, every regulatory deadline that matters for software and AI falls between 2025 and 2027. DORA has applied to financial entities since January 2025. NIS2 national laws were due in October 2024 and are still arriving unevenly across Member States. The EU AI Act's high-risk obligations land in December 2027, and Cyber Resilience Act reporting has been live since September 2026.
Infrastructure is the part most teams have already solved. The AWS European Sovereign Cloud has been generally available since January 2026, with EU-resident operations and customer content and metadata kept in the EU [2], and the EU AWS Regions have long supported data residency. What regulators now ask for goes further than where workloads run. They want evidence that every change to your systems was reviewed and approved, and that your AI systems are accurate, tested and overseen by people.
That evidence is what Connectory and Vectory produce. Connectory governs code before it merges. Vectory evaluates AI systems before they release. Both run on AWS and can be deployed in an EU AWS Region. This article takes each regulation in turn, quotes what it requires, and shows how the two products meet that requirement in practice.
What Each Regulation Requires From Software and AI Teams
| Regulation | Applies | Requirement that reaches your engineering teams |
|---|---|---|
| DORA | Since 17 Jan 2025, EU financial entities | Controlled change management, resilience testing including source code review, ownership of ICT third-party risk |
| NIS2 | National laws due 17 Oct 2024, essential and important entities | Secure development and vulnerability handling, supply chain security, management body accountability |
| EU AI Act | Prohibitions since Feb 2025, transparency since Aug 2026, high-risk from 2 Dec 2027 | Risk management, testing and validation, logging, human oversight, accuracy and robustness |
| Cyber Resilience Act | Reporting since 11 Sep 2026, full from 11 Dec 2027 | No known exploitable vulnerabilities, component inventory, 24h and 72h reporting |
Two points from the primary texts are worth holding on to. Under DORA, a financial entity that outsources ICT services remains fully responsible for compliance [6]. Under NIS2, management bodies must approve the cybersecurity measures, oversee them, and can be held liable for infringements [5]. Neither regulation lets you hand accountability to a cloud provider, which is why the evidence has to come from your own delivery process.
DORA: Controlled Change and Code Review
What it requires. DORA Article 9(4)(e) requires documented change management so that "all changes to ICT systems are recorded, tested, assessed, approved, implemented and verified in a controlled manner" [6]. Article 25 lists the tests a resilience testing programme should include, and "source code reviews where feasible" is on that list [6]. Article 5 makes the management body responsible for the ICT risk management framework [6].
How Connectory meets it. Connectory's SlopBuster reviews every pull request and reports its verdict as a GitHub check, with findings ranked as blockers or warnings [12]. Make that check required in branch protection and a change with unresolved blockers cannot merge. Each review produces a record of what changed, which policies it was checked against and what was found, and GitHub's own history records who approved and merged it. Together they form the "recorded, tested, assessed, approved" trail Article 9 describes, created automatically for every change rather than assembled before an audit.
Connectory's organisational memory holds your policies, architecture decisions and constraints, each tied to its source, and the reviewer checks every change against them. A DORA control such as "changes to payment services require a rollback plan" or "no new ICT third-party dependency without a register entry" becomes something checked on every pull request instead of something people have to remember.
How Vectory meets it. Many financial entities now run AI inside important functions: customer service agents, document processing, fraud triage. When those systems change, the change also has to be tested. Vectory's release gate runs an AI system against a versioned evaluation set and fails the pipeline if scores fall below threshold [15], which extends DORA-style controlled change to models, prompts and retrieval.
NIS2: Secure Development and Supply Chain Security
What it requires. NIS2 Article 21(2) requires risk-management measures including "supply chain security" and "security in network and information systems acquisition, development and maintenance, including vulnerability handling and disclosure" [5]. Member States must allow fines on essential entities of at least EUR 10 million or 2% of worldwide turnover, and EUR 7 million or 1.4% for important entities [5].
How Connectory meets it. Secure development needs evidence that security was checked during development, not only scanned afterwards. SlopBuster reviews each change for security issues, risky patterns common in AI-generated code, and conflicts with your own security policies before it merges [13]. For supply chain security, encode your dependency rules as Connectory policies, such as approved registries, licence rules, or a requirement to justify every new package, so every pull request that adds a dependency is checked against them. The per-change review record gives management bodies something concrete to oversee, which matters now that Article 20 makes them accountable [5].
AI coding assistants raise the stakes. Teams using Claude Code, Cursor, Codex or Copilot produce more code, faster, and more of it is written by tools that do not know your organisation's rules. Connectory works alongside those tools [12]. It does not replace them. It makes sure what they write meets your standards before it enters a regulated system.
How Vectory meets it. AI components are part of the system under NIS2 too. Vectory gives security and risk teams a repeatable way to test the AI parts of a service, including failure-mode analysis, before each release.
The EU AI Act: Testing, Oversight and Accuracy
What it requires. For high-risk AI systems, the AI Act requires a risk management system maintained across the lifecycle (Article 9), automatic logging (Article 12), effective human oversight (Article 14), and "an appropriate level of accuracy, robustness, and cybersecurity" maintained throughout the lifecycle (Article 15) [3]. Providers must run a quality management system that includes "examination, test and validation procedures" before, during and after development (Article 17) [3]. Deployers must keep system logs for at least six months (Article 26(6)) [3]. In July 2026 the digital omnibus, Regulation (EU) 2026/1744, moved these obligations to 2 December 2027 for Annex III systems such as credit scoring, employment and critical infrastructure, and to 2 August 2028 for AI in regulated products [4]. The requirements themselves did not change.
How Vectory meets it. Vectory is built for exactly this kind of testing [14].
- Accuracy and grounding (Articles 15 and 17). Vectory scores outputs with deterministic metrics, retrieval metrics such as precision, recall, MRR and NDCG, LLM-as-judge evaluation, and proof grounding, which checks that each claim in an answer is tied to cited evidence. Results are versioned, so you can show how accuracy held up across releases.
- Testing before release (Articles 9 and 17). The Vectory release gate fails a CI run when scores fall below a threshold or critical failure modes appear [15]. A model upgrade, prompt change or new retrieval source that degrades quality stops before production.
- Human oversight (Article 14). Human review annotations sit alongside automated scores, so reviewers record their judgement against the same evidence the gate uses. Pair that with a named approver in your release process, which Vectory supports but does not enforce on its own.
- Evidence for the technical file. Vectory exports a static report bundle, an HTML page plus JSON files covering scores, failure modes and a claim-evidence table, ready to attach to technical documentation and risk reviews [15].
Production logging stays with your runtime and observability stack, which is where Articles 12 and 26(6) place it.
How Connectory meets it. Article 15 also asks for cybersecurity, and AI systems are code. Connectory reviews every change to the application, the agent tools, the prompts kept in the repository and the infrastructure code, so the AI system's own software meets the same controls as everything else.
The Cyber Resilience Act: Shipping Without Known Vulnerabilities
What it requires. Since 11 September 2026, manufacturers of products with digital elements must report actively exploited vulnerabilities and severe incidents within 24 hours as an early warning and within 72 hours as a notification [8]. From 11 December 2027, Annex I requires products to be placed on the market "without known exploitable vulnerabilities", and requires manufacturers to identify and document vulnerabilities and components, including a software bill of materials [7].
How Connectory meets it. The cheapest vulnerability to report is the one that never ships. Connectory blocks changes with security findings before merge and checks dependency changes against your component policies, which supports both the "no known exploitable vulnerabilities" requirement and an accurate component inventory. When an incident does happen, the review history and GitHub's merge history together show which change introduced the affected code, what the review found, and who merged it, which shortens the investigation behind a 24-hour early warning.
How Vectory meets it. If your product includes AI features, its security and quality depend on model behaviour as well as code. Vectory's gate catches regressions in those features before each release.
Requirement-to-Capability Matrix
| Requirement | Source | Connectory | Vectory |
|---|---|---|---|
| Changes recorded, tested, assessed and approved | DORA Art. 9(4)(e) | Review record and required check on every pull request | Release gate on every AI change |
| Source code review in resilience testing | DORA Art. 25 | Automated review of every change against your policies | Not applicable |
| Secure development and vulnerability handling | NIS2 Art. 21(2)(e) | Security and policy findings block merge | Failure-mode analysis for AI components |
| Supply chain security | NIS2 Art. 21(2)(d) | Dependency and supplier rules checked as policies | Not applicable |
| Test and validation procedures | AI Act Art. 17 | Review of AI application and prompt changes | Versioned evaluation sets and CI gates |
| Accuracy, robustness and cybersecurity | AI Act Art. 15 | Security review of AI system code | Accuracy, grounding and retrieval scores per release |
| Human oversight | AI Act Art. 14 | Human merge decision with findings | Human review annotations alongside automated scores |
| No known exploitable vulnerabilities | CRA Annex I | Blocks changes with security findings | Regression gate for AI features |
Keeping It Sovereign: Deploying in an EU Region
Governance tools see your most sensitive material: source code, prompts, retrieved documents and model outputs. If they process it outside the EU, they undo the sovereignty work done at the infrastructure layer.
Connectory and Vectory both run on AWS and can be deployed in an EU AWS Region, configured to use models served within the EU, so code, review records, prompts and evaluation data stay in the EU. Vectory's deterministic metrics and proof grounding run wherever Vectory runs [15], and for LLM-as-judge scoring we configure judge models served within the EU. Connectory reviews repositories on GitHub.com. Our shared hosted Connectory service uses Standard Contractual Clauses for transfers outside the EEA [13], so customers with an in-EU processing requirement should use an EU deployment.
For workloads inside the AWS European Sovereign Cloud partition itself, the deciding factor is which models that partition offers for review and judging. AWS added in-Region inference for Gemma 4 open-weight models there in September 2026 [1], and we confirm model availability for each workload during scoping.
The UK and Wider EMEA
UK organisations face a parallel set of rules. The Cyber Security and Resilience Bill updates the UK NIS Regulations, brings managed service providers and data centres into scope, lets regulators designate critical suppliers, and sets incident reporting at 24 hours for an initial notification and 72 hours for a fuller report [9]. It was still before Parliament at the time of writing. In financial services, HM Treasury designated AWS, Google Cloud, Microsoft and Oracle as critical third parties, overseen by the Bank of England, the PRA and the FCA from 13 July 2026 [10]. That regulates the cloud providers, not the systems firms build on them. For AI, the UK has no AI Act, but the Data (Use and Access) Act 2025 requires information, the ability to challenge, and human intervention for significant automated decisions [11].
The same Connectory review trail and Vectory release evidence serve these UK requirements, and UK organisations that place AI systems on the EU market can also fall within the EU AI Act [3]. Across the wider EMEA region, including the Gulf states, data protection laws restrict cross-border transfers, which makes regional deployment of governance tooling just as relevant.
A 90-Day Path for AWS Customers
- 1Days 1 to 30: map obligations to controls. Identify which of DORA, NIS2, the AI Act, the CRA and the UK regimes apply to one priority workload. Write down the evidence each requires, using the matrix above as a starting point. Choose the EU Region or sovereign partition for the workload and for its governance tools.
- 2Days 31 to 60: switch on the gates. Deploy Connectory, load your security, architecture and regulatory policies, and make the review check required on production branches. Build a Vectory evaluation set from real, approved cases for each AI component and agree release thresholds with the business owner.
- 3Days 61 to 90: rehearse the audit. Take three recent changes and produce the full evidence chain for each: review record, evaluation report, human approver and production logs. Run a CRA-style drill to test a 24-hour early warning. Close whatever gaps you find.
Tactical Edge runs this as an engagement on AWS. Our agent governance practice designs the policies, approvals and audit trail, our AgentOps practice runs evaluation in production, and our AWS partnership covers Region and sovereign cloud architecture. To start with one workload, talk to our team.
Frequently Asked Questions
Does running on the AWS European Sovereign Cloud make us compliant?
No. It addresses where data and operations reside and who controls them [2]. DORA, NIS2, the AI Act and the CRA also require controlled change, secure development, testing and human oversight [6] [5] [3] [7]. Connectory and Vectory produce the evidence for those requirements.
The AI Act high-risk deadline moved to December 2027. Should we wait?
No. Building evaluation sets, thresholds and review trails takes months, and DORA, NIS2 and CRA reporting already apply [6] [5] [8]. The same controls produce evidence for all four regulations.
Can Connectory and Vectory run in the EU?
Yes. Both run on AWS and can be deployed in an EU AWS Region, with code, prompts, review records and evaluation data kept in the EU [13] [14]. For the sovereign partition, we confirm the available models for your workload first [1].
Do Connectory and Vectory replace our existing security and testing tools?
No. Keep your scanners, unit tests and penetration tests, which DORA Article 25 also lists [6]. Connectory adds policy-aware review of every change, including AI-generated code, and Vectory adds evaluation of AI behaviour, which traditional testing tools do not measure.
References
[1]AWS Security Blog. Run open-weight models on Amazon Bedrock in the AWS European Sovereign Cloud. https://aws.amazon.com/blogs/security/run-open-weight-models-on-aws-bedrock-in-aws-european-sovereign-cloud/
[2]AWS News Blog. Opening the AWS European Sovereign Cloud. https://aws.amazon.com/blogs/aws/opening-the-aws-european-sovereign-cloud/
[3]Regulation (EU) 2024/1689, the Artificial Intelligence Act. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
[4]Regulation (EU) 2026/1744, the Digital Omnibus on AI. https://eur-lex.europa.eu/eli/reg/2026/1744/oj
[5]Directive (EU) 2022/2555, the NIS2 Directive. https://eur-lex.europa.eu/eli/dir/2022/2555/oj
[6]Regulation (EU) 2022/2554, the Digital Operational Resilience Act. https://eur-lex.europa.eu/eli/reg/2022/2554/oj
[7]Regulation (EU) 2024/2847, the Cyber Resilience Act. https://eur-lex.europa.eu/eli/reg/2024/2847/oj
[8]European Commission. Cyber Resilience Act reporting obligations. https://digital-strategy.ec.europa.eu/en/policies/cra-reporting
[9]GOV.UK. Cyber Security and Resilience (Network and Information Systems) Bill: summary of the bill. https://www.gov.uk/government/publications/cyber-security-and-resilience-network-and-information-systems-bill-factsheets/summary-of-the-bill
[10]GOV.UK. UK financial system strengthened with new safeguards for major technology providers. https://www.gov.uk/government/news/uk-financial-system-strengthened-with-new-safeguards-for-major-technology-providers
[11]GOV.UK. Data (Use and Access) Act factsheet: UK GDPR and DPA. https://www.gov.uk/government/publications/data-use-and-access-act-2025-factsheets/data-use-and-access-act-factsheet-uk-gdpr-and-dpa
[12]Connectory. AI code governance platform. https://connectory.ai
[13]Connectory. Security and compliance. https://connectory.ai/security/
[14]Vectory. AI evaluation toolkit. https://vectoryai.com
[15]Vectory open-source repository on GitHub. https://github.com/tacticaledge/vectory