Has the AI Act been passed? Yes. The EU AI Act was adopted in 2024 and entered into force on 1 August 2024. Its obligations have phased in since February 2025. Enterprises now need to know which obligations apply today, which deadlines remain, and what evidence each deployed system requires. Aurascape supplies the runtime inventory and interaction records that evidence work depends on.

Last updated: September 2026.

Yes, the EU AI Act Is Law and Has Been in Force Since August 2024

The European Parliament adopted the text on 13 March 2024. The Council approved it on 21 May 2024. Publication in the Official Journal followed on 12 July 2024, as Regulation (EU) 2024/1689, and the Regulation entered into force twenty days later, on 1 August 2024 (EU AI Act, 2024). The EU AI Act is binding law. Its individual obligations apply on the dates set in the Regulation’s final provisions.

The EU AI Act is Regulation (EU) 2024/1689, a directly applicable European law that governs how AI systems are placed on the EU market and used inside it, with duties tiered by the risk a given use creates rather than by the technology behind it. The Regulation applies directly across EU member states. National authorities run enforcement and procedure inside the framework it sets, so member states do not write their own version of the substantive duties.

The reach extends past Europe. The Act covers providers that place AI systems on the EU market, deployers established in the EU, and providers or deployers based in third countries where the output produced by the system is used in the EU (European Commission, 2026). A company established outside the EU falls within scope when it places an AI system on the EU market or when the system’s output is used in the EU. Whether it acts as a provider or a deployer depends on how it supplies or uses that system, and that turns on the facts of each case.

What Is the Difference Between Entry Into Force and Application?

Most of the confusion about whether the AI Act has passed comes from mixing two milestones. Entry into force is the date the law legally exists: 1 August 2024. Application is the date a specific obligation starts to bite, and the Regulation staggers those dates across several years so that market surveillance structures, harmonized standards, and vendor conformity work can catch up. A page that calls the Act pending or proposed is wrong. So is a page that says every duty landed in August 2024. The law is in force, and its obligations arrive on a published schedule.

What EU AI Act Obligations Apply Right Now?

The Regulation phases its obligations across several application and transition dates. Prohibited practices and the AI literacy duty applied from 2 February 2025. Obligations for general purpose AI models, the governance structure, notifying authorities, and most penalty provisions applied from 2 August 2025. The Regulation reached general application on 2 August 2026 (European Parliament, 2025).

Enterprises often misread the August 2025 milestone. Its model obligations fall on providers of general purpose AI models, who must hold technical documentation, publish a sufficiently detailed summary of the content used for training, and operate a copyright policy, with additional duties where a model presents systemic risk. An organization that buys, embeds, or fine-tunes such a model is usually not the model provider. What that milestone gives a deployer is a documentation requirement to point to: the documentation a model provider must now maintain is documentation you can request, and reference in your own records, before a system goes live.

Confirm the high-risk timing before planning against it. Article 113 as enacted set application for stand-alone Annex III high-risk systems at 2 August 2026, and for AI acting as a safety component in products already covered by Annex I product legislation at 2 August 2027. Those high-risk dates have since been the subject of amending legislation, so check the consolidated text before committing a program plan to a specific high-risk deadline.

A separate transition provision, Article 111, covers AI components of the large-scale EU information technology systems listed in Annex X that were placed on the market before the high-risk application date. Those components run to 31 December 2030. The AI Act timeline below separates the milestones that have already arrived from the high-risk dates that need confirmation.

Two things follow for a program running today. First, the prohibition, AI literacy, general purpose model, and transparency milestones are live commitments, not preparation, so an internal plan that files them under future work is already behind. Second, the high-risk workstream is preparation whose deadline you confirm rather than assume, and the preparation itself does not change: classify the use, fix your role, assign human oversight, and start producing records of real use now, so the evidence exists whenever the date lands.

What Risk Tier Applies, and Who Owns the Obligation?

The Act structures duties by risk. Two of the labels below, prohibited and high-risk, are terms the Regulation itself uses. The transparency grouping and the residual category are common shorthand for the rest of the structure, not additional statutory tiers.

• Prohibited practices (Article 5). Banned since 2 February 2025. The article covers practices such as social scoring, untargeted scraping of facial images to build recognition databases, emotion inference in the workplace or education outside narrow safety and medical exceptions, and certain manipulative or predictive-policing uses.
• High risk. The Annex III use cases such as employment and worker management, education, creditworthiness, essential services, biometrics, critical infrastructure, law enforcement, and migration, plus AI acting as a safety component in Annex I regulated products. Risk management, data governance, logging, human oversight, accuracy, and conformity assessment all attach here.
• Transparency obligations (Article 50). These split by role. Providers must design systems that interact directly with people so it is clear a person is dealing with AI, and providers of systems generating synthetic audio, image, video, or text must mark those outputs in a machine-readable format. Deployers carry their own disclosure duties, including for emotion recognition, biometric categorisation, and published deepfake content. The specific duty follows the role and the system, not one universal rule.
• Other AI systems. Systems that are not high risk can still carry AI literacy, transparency, role-specific, and sector requirements, and a change in how a system is used can move it into a different category.

General purpose AI models sit on their own track, as covered above. Most enterprises act as deployers, using a system under their own authority rather than building one and placing it on the market. That role is not fixed. A deployer that puts its own name or trademark on a high-risk system, substantially modifies it, or changes its intended purpose so that it becomes high risk takes on provider obligations as well.

Enforcement runs through national market surveillance authorities, with the AI Office overseeing general purpose AI models. Article 99 sets the ceilings. Breach of the prohibited practices carries up to EUR 35 million or 7% of total worldwide annual turnover for the preceding financial year, whichever is higher. Most other breaches, including deployer duties, carry up to EUR 15 million or 3%. Supplying incorrect, incomplete, or misleading information to notified bodies or national authorities carries up to EUR 7.5 million or 1%. For small and medium enterprises, the applicable maximum is capped according to the lower amount specified by Article 99.

Deployer Obligations Are an Evidence Problem, Not a Paperwork Problem

Article 26 sets out what a deployer of a high-risk system must do: use the system in line with the provider’s instructions, assign human oversight to people with the competence, training, and authority to exercise it, ensure input data is relevant and sufficiently representative for the intended purpose where the deployer controls that data, monitor operation, notify the provider and the relevant authorities of risks and serious incidents, inform workers and their representatives before putting such a system into service at work, and keep the logs the system generates automatically for a period appropriate to the intended purpose, at least six months unless other law provides otherwise.

Provider documentation and conformity work describe the system and its assessed design. Runtime records show how people and agents used it after deployment. Those answer different evidence questions, and a compliance program needs both. Without runtime records, a team has nothing to show for how the system behaved in production, which is exactly the question an oversight duty invites.

Scope determination has the same shape. An organization cannot tier a risk it has not first found, and the surface keeps moving. Aurascape has detected that 50+ new AI apps release every day, and AI features keep arriving inside SaaS products employees already license (Aurascape Discover and Monitor AI, 2026). Sector rules layer on top of the Act rather than replacing it, which is why the mapping work differs across enterprise AI programs, financial services, healthcare and pharmaceutical organizations, and education institutions.

How to Build EU AI Act Evidence From Live AI Usage

Work each deadline backwards from the artifact an auditor will ask for. This sequence produces evidence as a by-product of normal operation, instead of a documentation sprint before every milestone.

1. Inventory every AI system, account, and agent in use, sanctioned and unsanctioned, including AI features inside licensed SaaS and assistants running on developer machines.
2. Classify each use against the risk tiers, and write down the reasoning, not just the conclusion.
3. Fix your role per system, provider, deployer, or both, and re-check it whenever a team rebrands or repurposes a tool.
4. Assign named human oversight for anything high risk, and record who holds it and what authority they have to intervene.
5. Instrument the interaction rather than the perimeter: prompts, files, code, responses, account type, tool calls, and agent actions.
6. Enforce in a way that creates a decision record. The available actions are allow, coach, notify, redact, redirect, and block.
7. Govern access to those records through role-based access control, align record handling with applicable requirements, and rehearse the audit question: show one week of use by user, data category, and policy decision.

Aurascape’s discovery output feeds steps one and two, and its runtime controls support steps five through seven. Classification and role determination under the Act stay with compliance and legal teams. Aurascape supplies the inventory and the interaction evidence those judgments rest on.

Discovery runs continuously across network, endpoint, and API planes against a catalog of 30,000+ AI apps and agents, and an endpoint agent finds AI running locally on laptops and servers (Aurascape, 2026). Each discovered application carries a risk profile covering category, model and mode, entitlement tier, and 25+ risk attributes, so a classification exercise starts from something concrete.

Inline, the Aurascape platform decodes the AI exchange rather than the destination: prompts and responses, files and artifacts, code, identity and entitlement, whether the session runs on an enterprise tenant or a personal account, and application-specific Intentions such as summarize, upload, generate code, or agent mode. Real-time data protection applies 600+ data classifiers to that traffic, so the record of sensitive data handling sits at the interaction, not in a policy document. Enforcement happens before data reaches the AI service, and every action taken becomes part of the record. Aurascape is additive to an existing secure service edge (SSE), cloud access security broker (CASB), data loss prevention (DLP), or secure web gateway (SWG) stack, with no rip and replace.

Agent activity needs the same treatment on the execution side. Aurascape discovers and secures local AI agents and their interactions, and adds a Zero-Bypass MCP Gateway that marks every tool call it approves and blocks unmarked calls, governing the agent-to-tool execution path inline rather than observing it (Aurascape Secure Agentic AI, 2026). Model Context Protocol (MCP) is one common tool-execution pattern, not the whole agent access-control problem. Aurascape creates interaction records for audit and effectiveness, governed by role-based access control (RBAC) for privacy, and for MCP workflows those records can include the tool invoked and the policy decision applied. Auri gives compliance, legal, and business teams role-based, natural-language access to that AI usage, risk, and policy information without a security console or a query language, so the compliance owner answers a regulator question without filing a ticket.

Deployer evidence draws on two complementary sources: documentation prepared before deployment, and runtime records that show actual use. The table below sets out what each contributes.

In one Aurascape deployment, The Police Credit Union built a two-phase framework for adopting AI responsibly, visibility first and protection second, with projected results of 27% productivity gains and 83% risk reduction as published in the Aurascape case study (Aurascape customer case study, The Police Credit Union, 2026).

Frequently Asked Questions

Has the AI Act been passed?

Yes. The EU AI Act was adopted in 2024, published as Regulation (EU) 2024/1689, and entered into force on 1 August 2024. It is binding law across the EU, and its obligations apply on a staggered schedule rather than all at once.

When did the AI Act pass?

The European Parliament adopted the text on 13 March 2024 and the Council approved it on 21 May 2024. Publication in the Official Journal followed on 12 July 2024, and entry into force was 1 August 2024.

Is the EU AI Act in effect now?

Yes, in stages. Prohibited practices and AI literacy duties have applied since 2 February 2025, general purpose AI model obligations, governance, and most penalty provisions since 2 August 2025, and the Regulation reached general application on 2 August 2026.

What still has a future deadline?

The high-risk tier is the part to confirm. Article 113 as enacted set 2 August 2026 for stand-alone Annex III systems and 2 August 2027 for AI embedded in Annex I regulated products, and those dates have since been the subject of amending legislation, so check the consolidated text before planning against either. A separate Article 111 transition for AI components of the Annex X large-scale EU information technology systems runs to 31 December 2030.

Does the EU AI Act apply to companies outside the EU?

Yes, where the AI output is used in the EU or the system is placed on the EU market. A company headquartered elsewhere can be a deployer or provider under the Act depending on how it supplies or uses the system.

What are the penalties under the EU AI Act?

Article 99 sets maximum fines: up to EUR 35 million or 7% of total worldwide annual turnover, whichever is higher, for prohibited practices; up to EUR 15 million or 3% for most other breaches, including deployer duties; and up to EUR 7.5 million or 1% for supplying incorrect or misleading information to authorities. For small and medium enterprises, the applicable maximum is capped according to the lower amount specified by Article 99.

What is the difference between a provider and a deployer?

A provider develops an AI system and places it on the market, while a deployer uses it under its own authority. Enterprises commonly act as deployers, and a deployer becomes a provider if it puts its own name on a high-risk system, substantially modifies it, or changes its intended purpose.

Is there a US AI Act?

Searches for an AI Act usually mean the EU Regulation. In the United States, AI obligations currently arrive through state legislation and existing sector regulators rather than through one directly applicable federal regulation of the same scope. Colorado’s SB 24-205 is the state law most often compared with the EU approach, because it tiers duties for developers and deployers of high-risk systems, and its operative date has been amended (Colorado General Assembly, 2024).

How do deployers evidence compliance after go-live?

With records of real usage alongside provider documentation: which AI systems and agents were used, by whom, under which account, what data moved, which tools an agent invoked, and what policy decision applied. Interaction-level records help deployers answer those questions and support the wider evidence program a high-risk system requires.

Aurascape gives compliance, legal, and security teams the runtime inventory and interaction records that EU AI Act evidence work needs: what AI is in use, who used it, which account, what data moved, which tool an agent invoked, and what policy decision followed. Documentation describes intent. Interaction records show practice. A short demo walks through how that evidence is produced from everyday AI use and mapped to deployer duties.

See how Aurascape turns live AI usage into EU AI Act evidence →