Artificial
intelligence
The AI Act, risk classification, obligations for providers and deployers, and AI in contracts. Gathered in one place and kept up to date.
When the AI Act applies
The Regulation applies in stages. The dates below apply in the EU. When the rules apply in Norway depends on incorporation into the EEA Agreement.
- 1 Aug 2024 The Regulation enters into force
- 2 Feb 2025 Prohibited practices and AI literacy requirements
- 2 Aug 2025 General-purpose AI models and governance
- 2 Aug 2026 The bulk of the rules, including high-risk systems under Annex III
- 2 Aug 2027 High-risk systems linked to product legislation, Art. 6(1)
The EU has proposed postponing certain high-risk rules. The timeline will be updated as the process develops.
Articles on this topic
AI in recruitment: what does the GDPR require before the EDPB's 2027 enforcement action?
On 11 November 2026 the EDPB announced that automated decision-making and AI in recruitment will be the topic of the coordinated enforcement action starting in 2027. Employers that let a tool filter out applicants without a human assessing them take a decision under GDPR Art. 22, which requires an exception, human intervention and an explanation of the logic to the applicant.
Fundamental rights impact assessment: who must carry one out under the AI Act?
Under Article 27 of the AI Act, a fundamental rights impact assessment (FRIA) must be carried out before the first use of a high-risk system by bodies governed by public law, private entities providing public services, and deployers using AI for creditworthiness assessment or for risk assessment and pricing in life and health insurance. The obligation applies in the EU from 2 December 2027.
Cloud suspension clauses: when may the provider shut down an AI service?
Cloud suspension clauses in the standard terms of OpenAI, Anthropic, AWS, Google and Microsoft allow the provider to suspend the service immediately for policy breaches, security risks, non-payment and legal prohibitions, and to terminate the agreement. The customer rarely has a claim for damages. Notice, proportionate suspension, a wind-down period and data export must therefore be negotiated.
OpenAI intermediary approval: what does it mean for customers of AI platforms?
From 20 November 2026, OpenAI's Usage Policies require prior approval for intermediaries that give third parties direct access to OpenAI models, such as model routers and AI gateways. Applications that merely integrate the models into their own features are not covered. An intermediary without approval risks suspension, and the risk passes down the contract chain to its customers.
Reselling AI services: when does a business become a provider under the AI Act?
When reselling AI, a business is the provider under Article 3(3) of the AI Act where it offers a system it has developed or had developed under its own name. Under Article 25(1), distributors, importers and deployers also become the provider of a high-risk system where they put their name on it, substantially modify it or change its purpose so that it becomes high-risk.
Prompt injection: which security requirements should an AI agent contract contain?
Prompt injection is an attack in which hidden instructions in text make a language model act against the business. The risk cannot be eliminated, so it must be contained through least privilege, human approval and logging. The contract should bind the vendor to concrete security measures, testing, notification of vulnerabilities and incidents, and liability proportionate to the risk.
Text and data mining in Norway: a new exception, and what are AI vendor indemnities worth?
The Norwegian Copyright Act currently has no text and data mining exception. Prop. 41 LS (2025–2026) proposes a new section 50 e permitting text and data mining of lawfully accessible works unless the rightholder has reserved its rights in an appropriate manner. Foreign case law is divided, and AI vendor indemnities carry broad exclusions and Irish governing law.
AI agent governance: what do the GDPR and the Norwegian DPA require?
AI agent governance is part of the controller's duty to implement appropriate technical and organisational measures under GDPR Articles 24, 25 and 32. On 28 October 2026 the Norwegian Data Protection Authority advised organisations to map their agents, restrict what the agents can access and do, give each agent its own identity and appoint a person responsible who can stop it.
Press publishers' right and AI chatbots: what does the Advocate General say in C-250/25?
The press publishers' right under Article 15 of the DSM Directive may be infringed when an AI chatbot retrieves a press article in real time and reuses its expression in an answer, according to Advocate General Szpunar in C-250/25 Like Company. Decisive are unlawful access and whether the answer replaces the original. Training is, in his view, as a rule permitted under the TDM exception.
Meta's new terms: what does the injunction granted to Norwegian publishers mean?
The Oslo District Court has by interim injunction prohibited Meta from introducing Meta's new terms for business accounts against ten Norwegian publishers until their lawfulness is finally decided. The order was made without hearing Meta, which may demand a subsequent oral hearing under Section 32-8 of the Dispute Act. The injunction protects only the applicants.
The complaints against Scale AI: which rules govern the monitoring of platform workers?
On 27 October 2026, noyb filed five complaints against Scale AI with the supervisory authorities in Austria, France, Italy and the Netherlands. They concern tracking tools, biometric checks of annotators on the Outlier platform and unanswered access requests. As Scale AI has no main establishment in the EEA, each authority may handle its complaint itself, as may the Norwegian authority.
Chinese AI models: how does DeepSeek as an app differ from locally run model weights?
The risk of Chinese AI models depends on how they are used. The DeepSeek app and DeepSeek's API store personal data in China, which has no adequacy decision, so their use requires a transfer mechanism under Article 46 GDPR. Where the open model weights are run on in-house infrastructure or by a provider within the EEA, no transfer to China takes place, although supply chain risk remains.
Standard terms by reference: what does the Rototec judgment mean for liability caps?
Under Norwegian law, a reference to standard terms in a tender does not automatically make their limitations of liability part of the contract. In HR-2026-780-A (Rototec), the Supreme Court held unanimously that the reach of such a reference depends on its wording, presentation and placement, and that any ambiguity falls on the party that drafted the tender.
Digital Omnibus: what the Council mandate means for AI training, cookies and breaches
The Council's Digital Omnibus mandate was agreed in Coreper on 21 October 2026 and is not adopted law. The Council keeps the definition of personal data, allows AI development on the basis of legitimate interest, supports a 96-hour breach notification deadline and keeps the cookie rules in the ePrivacy Directive. Current law continues to apply in full in Norway.
Blocking AI agents: can an online shop shut them out? Amazon v Perplexity
Blocking AI agents through terms of use and technical barriers is open to an online shop. In Amazon v Perplexity, however, the US Ninth Circuit held on 4 August 2026 that it is the user who obtains access when the agent acts. Under Norwegian law the agent's purchase normally binds the consumer, while the risk of card misuse follows Financial Contracts Act §§ 4-2 and 4-30.
Data annotator monitoring: what should buyers require of AI vendors?
Data annotator monitoring is primarily the vendor's responsibility under the GDPR, also where annotators are contractors. The buyer may still incur liability through its own data, joint controllership and the Norwegian Transparency Act. Contracts should require a list of monitoring tools, a ban on advertising trackers, limited biometrics, human decisions on deactivation and audit rights.
EN 18286 and ISO/IEC 42001: which gives a presumption of conformity under the AI Act?
EN 18286:2026 is the first harmonised standard under the AI Act. Its reference was published in the Official Journal on 14 October 2026, and providers of high-risk AI systems that apply it are presumed to comply with the quality management system requirements in Article 17 to the extent the standard covers them. ISO/IEC 42001 certification gives no such presumption.
General-purpose AI models: what must the provider do, and what should the customer demand?
Under Article 53 of the AI Act, providers of general-purpose AI models must keep technical documentation, inform downstream providers, adopt a copyright policy and publish a summary of their training content. From 2 August 2026 the Commission may impose fines of up to 3 per cent of worldwide turnover. Norwegian customers should secure warranties, indemnities and access to documentation.
AI in central government 2027–2029: what does the new common directive require?
From 2027, all Norwegian government agencies must work for greater efficiency and quality with particular emphasis on the responsible use of AI in central government. The requirement follows from circular D-3/2026 and the letters of allocation for 2027–2029. Each agency sets its level of ambition with the employee representatives and reports on measures and results in its annual report.
AI meeting minutes: what does the GDPR require for recordings and transcripts?
AI meeting minutes involve recording, transcription and draft minutes, and each step requires a legal basis under GDPR Art. 6. Datatilsynet recommends deleting the audio once the transcript exists, and the transcript once the minutes are final. Private businesses will normally rely on legitimate interests, must inform all participants before recording and should as a rule carry out a DPIA.
AI vendor terms for high-risk use: what do Anthropic, OpenAI, Google and Microsoft demand?
From 12 November 2026, Anthropic's Usage Policy requires a qualified person with authority to change the output to review AI recommendations on employment, credit, insurance and healthcare, and the affected person to be told that AI was used. OpenAI, Google and Microsoft have similar AI vendor terms. A breach may lead to suspended access, even though the AI Act does not yet apply in Norway.
Can email, Slack and source code be sold as training data on winding-up or bankruptcy?
Email, Slack and source code may only be sold as training data where each category of data has a lawful basis. Personal data requires a legal basis and a compatible purpose under the GDPR, employees' mailboxes must as a rule be erased, and customer data processed as a processor cannot be sold. Code and customers' confidential information are bound by copyright, licences and confidentiality.
Automated decisions: what does Uber's EUR 825 million fine mean for Norwegian companies?
The Dutch data protection authority has fined Uber EUR 824.99 million because driver accounts were deactivated automatically without a human assessing the case, and because drivers were given too little information. Automated decisions with significant effects on individuals are prohibited as a starting point under GDPR Art. 22. Uber has appealed the fine.
AI agents and contracting: is the business bound by what the agent agrees?
Contracts concluded by AI agents bind, as a rule, the business that put the agent to work. Norwegian law has no specific rules on this, but the agent's order is probably treated as the business's own declaration. Exceptions may apply where the counterparty understood or ought to have understood that something was wrong. System limits and agreed ground rules manage the risk best.
AI content labelling: what does Article 50 of the AI Act require?
AI content labelling is required in the EU from 2 August 2026 under Article 50 of the AI Act. The provider of the tool must mark the content in a machine-readable format. The business using the tool must visibly disclose deepfakes and AI-generated text on matters of public interest. The regulation does not yet apply in Norway, but Norwegian businesses publishing in the EU may be caught.
AI smart glasses: what is allowed today, and what does the announced ban mean?
AI smart glasses are not banned in Norway today. If the wearer films other people beyond purely private use, the GDPR applies, and facial recognition in practice requires explicit consent. Use at work is a control measure under the Working Environment Act. On 5 October 2026 the government announced a temporary ban covering public places among others, but no bill has been put forward yet.
AI agent liability: who pays when an AI agent causes harm?
Under Norwegian law, AI agent liability rests primarily with the business that puts the agent to work. An agent is neither a legal person nor an employee, so the harm is treated as the business's own. Towards customers, ordinary contractual liability applies. The vendor is liable only as far as the contract allows, and the new EU Product Liability Directive does not yet apply in Norway.
Anonymisation under the EDPB guidelines: is your data really anonymous?
Anonymisation under the GDPR requires the likelihood of identifying anyone to be insignificant, assessed from the perspective of whoever will use the data. The draft EDPB guidelines 02/2026 build on the SRB judgment and apply three criteria, singling out, linkability and inference. A processor inherits its customer's perspective, and the assessment must be repeated over time.
Checking AI-generated text: what can businesses learn from a ruling set aside?
Checking AI-generated text means checking it against the source, not against how plausible it sounds. In September 2026 a Norwegian court of appeal set aside a ruling that set out nine paragraphs of submissions no party had made, and the court has confirmed AI was used. The same error in contracts, board papers or replies to regulators binds the business or misleads the board.
Norway's AI agreement: what must employers discuss before introducing AI?
Norway's AI agreement of 2 October 2026 is a letter of intent between the government and the social partners. It places no new obligations on individual employers, whose duties come from the Working Environment Act and collective agreements. If AI changes how work is organised, or can be used to monitor staff, it must be discussed with employee representatives before the decision is made.
Personal data in test environments: can we test with real customer data?
Personal data in test environments may only be used where testing is compatible with the purpose for which the data was collected, under GDPR Art. 6(4), and where synthetic or pseudonymised data will not do. Private businesses have no special rule. Copies of production data used for testing need the same security, access control and deletion as production.
AI hallucinations: who is liable when your chatbot promises a customer too much?
A business is liable for AI hallucinations in its own customer service. The chatbot is not a separate legal person, and what it says counts as the business's own statements. Towards consumers, wrong answers can give rise to claims for non-conformity, and misleading answers can breach Norwegian marketing law. A disclaimer in the chat window helps little.
AI literacy is now a requirement: what Article 4 of the AI Act means for your business
AI literacy is a requirement under Article 4 of the AI Act. Businesses that use AI at work must take measures to build literacy among staff and others who use the systems on their behalf. The requirement has applied in the EU since 2 February 2025 and was softened in July 2026. It does not yet apply in Norway, but should be prepared for now.
AI policy for employees: what can an employer require and monitor?
Under Norwegian law, an employer can set an AI policy for employees under its managerial prerogative, including bans on certain tools and requirements to check the output. Monitoring how employees actually use AI is a control measure that requires objective grounds, discussion with employee representatives and prior information. Continuous monitoring of individual use is generally prohibited.
Board responsibility for AI: what should the board ask before its next meeting?
Board responsibility for AI follows from the general rules of the Norwegian Companies Act. The board must ensure the business is properly organised, subject to adequate control and that the chief executive is supervised, and this also covers the use of artificial intelligence. The board does not need technical expertise, but it must know where AI is used, what the risk is and who is responsible.
Copyright and AI: who owns the text, code and images that AI creates?
Content that AI creates entirely on its own probably has no copyright under Norwegian law, because the Copyright Act requires original and individual creative effort by a human. If a human reworks the output creatively, the human contribution may be protected. A vendor's promise that you own the output gives no more protection than the law itself provides.
DPIA for AI tools: when is it required, and what must it contain?
A DPIA for AI tools is required when their use is likely to result in a high risk to data subjects, under GDPR Art. 35. AI tools with access to email, documents, customer data or employee data often meet the threshold. If two or more of the nine criteria in the Article 29 Working Party guidelines are met, you should assume that a DPIA is required.
SaaS limitation of liability: what holds up under Norwegian law?
A SaaS limitation of liability is, as a rule, valid between businesses under Norwegian law. A liability cap and an exclusion of indirect loss normally hold, but as a starting point they do not protect the supplier against wilful misconduct or gross negligence. Unusual and onerous terms may also fall away as not accepted, be read restrictively or be adjusted under the Contracts Act § 36.
Your secrets in ChatGPT: can you lose protection for trade secrets?
Trade secrets are only protected if the business has taken reasonable steps to keep them secret. One employee pasting a confidential document into ChatGPT does not normally remove that protection. But if the business lets employees use AI tools freely, without rules and without an enterprise agreement, it becomes hard to show reasonable steps when the secret later has to be enforced.
Vibe lawyering: what is the risk when management drafts contracts with AI?
Vibe lawyering means managers and specialists producing contracts with AI without legal review. The contract binds the company in full, even when it rests on foreign concepts, hallucinated references or liability rules that do not work under Norwegian law. AI can be used for simple agreements, but not without review when a lot is at stake.
The legal framework
- General Data Protection Regulation (EU) 2016/679EU
- Norwegian Personal Data Act (personopplysningsloven)lov
- Norwegian Equality and Anti-Discrimination Act (likestillings- og diskrimineringsloven)lov
- AI Act (EU) 2024/1689EU
- Regulation (EU) 2026/1744 (digital omnibus on AI)EU
- Data Act, Regulation (EU) 2023/2854EU
- Council Regulation (EU) No 833/2014 concerning restrictive measures in view of Russia's actions (consolidated 24 July 2026)EU
- Norwegian Sanctions Regulation concerning Ukraine (sanksjonsforskrift Ukraina)forskrift
- DORA, Regulation (EU) 2022/2554 on digital operational resilience for the financial sectorEU
- Norwegian DORA Act (DORA-loven)lov
- Norwegian Contracts Act (avtaleloven)lov
- Cyber Resilience Act, Regulation (EU) 2024/2847EU
- Norwegian Digital Security Act (digitalsikkerhetsloven)lov
- Digital Single Market Copyright Directive (EU) 2019/790EU
- Norwegian Copyright Act (åndsverkloven, in Norwegian)lov
- Directive 2001/29/EC on copyright in the information society (InfoSoc Directive)EU
- Norwegian Dispute Act (tvisteloven)lov
- Norwegian Competition Act (konkurranseloven)lov
- Regulation (EU) 2022/1925 on digital markets (Digital Markets Act)EU
- Platform Work Directive (EU) 2024/2831EU
- Norwegian Electronic Communications Actlov
- ePrivacy Directive 2002/58/ECEU
- NIS2 Directive (EU) 2022/2555EU
- Norwegian Cancellation Act (angrerettloven)lov
- Norwegian E-Commerce Act (ehandelsloven)lov
- Norwegian Financial Contracts Act (finansavtaleloven)lov
- Norwegian Penal Code (straffeloven)lov
- Norwegian Marketing Control Act (markedsføringsloven)lov
- Norwegian Working Environment Act (arbeidsmiljøloven)lov
- Norwegian Regulations on employers' access to email accounts and other electronically stored material (e-postforskriften)forskrift
- Norwegian Transparency Act (åpenhetsloven)lov
- Commission Implementing Decision (EU) 2026/2274 on the harmonised standards for high-risk AI systemsEU
- Regulation (EU) No 1025/2012 on European standardisationEU
- Norwegian Public Administration Act 2025 (Act of 20 June 2025 No. 81, not yet in force)lov
- General Data Protection Regulation (EU) 2016/679EU
- Norwegian Child Welfare Act (barnevernsloven)lov
- Norwegian Private Limited Liability Companies Act (aksjeloven)lov
- Norwegian Bankruptcy Act (konkursloven)lov
- Norwegian Creditors Recovery Act (dekningsloven)lov
- Norwegian Trade Secrets Act (forretningshemmelighetsloven)lov
- Platform Work Directive (EU) 2024/2831EU
- Norwegian Regulations on camera surveillance in businesses (forskrift om kameraovervåking i virksomhet)forskrift
- Norwegian Damages Act (skadeserstatningsloven)lov
- Norwegian Sale of Goods Act (kjøpsloven)lov
- Norwegian Product Liability Act (produktansvarsloven)lov
- Product Liability Directive (EU) 2024/2853EU
- Norwegian Code of Conduct for Lawyers (Regler for god advokatskikk), annex 1 to the Lawyers Regulationsforskrift
- Norwegian Financial Supervision Act (finanstilsynsloven)lov
- Norwegian Consumer Purchases Act (forbrukerkjøpsloven)lov