Many companies are baking artificial intelligence into their products, including, with our help, several Daedalus clients. Whether it’s sunglasses hooked up to a chatbot, an autonomous vehicle, or a camera that can follow its subject, AI is showing up in all kinds of products. If you’re building a new product that uses AI, you may have gotten used to building with few or no additional AI regulations or standards affecting your development process. Now, however, if you plan on selling your products in Europe, a new set of regulations apply. The EU Artificial Intelligence Act (Regulation (EU) 2024/1689) was introduced in 2024 and remains one of the first comprehensive legal frameworks for AI anywhere in the world. Since its inception, it has been rolling out sweeping new AI regulations for all products on the European market, and many of its most impactful rulings are already in effect.
The first thing to know about the AI Act is that it largely categorizes its regulations based on use cases rather than any specific technologies. Each device is sorted into one or more of four tiers based on the risk its specific application poses: prohibited practices, high-risk systems, systems subject to transparency requirements, and minimal- or no-risk systems. Notably, most AI systems fall into the last category, which is not subject to many of the new obligations. This applies to anyone selling products in Europe, regardless of where the company is based.
When the AI Act was introduced in 2024, it launched a timeline to release new regulations in order to give companies time to prepare.
In February 2025, eight categories of AI practice became prohibited outright in the EU. These mostly included government applications like social scoring and predictive policing, but it also banned some commercial and industrial applications like inferring emotions in workplaces and schools or surveillance of employee attention rates through AI cameras. The ban covers AI inferences from sensitive biometric data such as race or sexual orientation in public spaces. The act does, however, carve out specific exceptions for medical and safety applications. For example, a device to detect drowsiness in truck drivers to prevent accidents is fine, but a device that measures employee productivity through microexpressions on employees’ faces is not.
In August 2025, general-purpose artificial intelligence (GPAI) makers became subject to new regulations governing their unreleased foundation models. Companies like OpenAI, Meta, and Mistral were required to release technical documentation, a copyright policy, and a public summary of what data was used to train their new models. Most companies are not directly providing access to general-purpose models, but if you’re providing chatbots to users or relying on these services, it’s probably worth looking into these regulations.
Originally, most other regulations were scheduled to go into effect in August of 2026. If your product is considered to be in the high-risk tier, you would have been required to create and release extensive documentation by this date. The high-risk tier includes systems in areas like biometrics, critical infrastructure, employment, safety components, or other essential services. These products would require a documented risk management system, data governance, technical documentation, logging, human oversight measures for deployment and use, accuracy and cybersecurity testing, a conformity assessment, CE marking, and registration in a special EU database. Needless to say, this will mean additional responsibilities for product designers and engineers.
New regulations for the lower-risk transparency tier have gone into effect. Certain systems must disclose AI interactions or AI-generated content under specific transparency rules. For example, disclosures that a user is interacting with an AI when speaking to a chatbot are required when that is not otherwise obvious.
Leading up to the deadline, however, the timeline changed. The EU was unable to put together harmonized standards ahead of initial target dates. The harmonized standards are the technical specifications that allow a manufacturer to demonstrate compliance by following a manual instead of having to argue an interpretation of compliance before a regulator. Initially, the European standards bodies CEN and CENELEC were supposed to deliver them by August 2025, but that process ended up being pushed out to late 2026. To make matters worse, as of March 2026, only 8 of the 27 EU member states had appointed their enforcement authorities. The Commission also missed its own deadline for publishing classification guidance for determining a product’s risk category. This would have meant that the rules would’ve gone into effect before companies had proper guidance on how to comply with them.
To address this, the EU enacted a simplification package in June 2026 known as the “AI Omnibus” (Regulation (EU) 2026/1744), which formally went into force in July 2026. This new package pushed out the timeline for several regulations and created easier-to-follow guidelines for the upcoming dates. The most pressing note is that the transparency tier deadline of August 2, 2026, did not change. Products marketed in the EU must meet the disclosure requirements applicable to their specific AI uses. If you’re selling an AI-enabled product in the EU, this is something you absolutely should be paying attention to when designing your product.
An extension of the original August 2026 deadline has also been granted for embedded AI devices. Embedded AI in medical devices, machinery, vehicles, and other regulated devices had already been subject to conformity assessments under existing regulations. Because of all of this, qualifying embedded devices have been given until August 2028 to comply with the new regulations. The high-risk tier’s regulations have also been further clarified and extended. The first big change is for machinery products. The regulation for these products has been streamlined to minimize regulatory overlap between the AI Act and existing machinery regulatory bodies. This makes it easier for machinery manufacturers to keep their products compliant and eliminates redundant effort for regulators. Additionally, the simplification package clarified that AI technology that assists users or optimizes performance is not automatically considered to be high-risk if its failure would not create a safety risk for users.
As for the specific deadlines, the first one to watch is December 2, 2026. This date marks the end of the transitional grace period for certain existing generative AI systems. By December 2, providers of covered tools placed on the market before August 2, 2026, must ensure their outputs are marked in a machine-readable form so that algorithms can detect new AI-generated content. Newly marketed systems covered by this requirement must comply from the outset. High-risk requirements for standalone systems have now been pushed back to December 2, 2027, and high-risk requirements for AI in regulated embedded systems are pushed back to August 2, 2028. While the first official harmonized standards have finally begun releasing, product designers and engineers should keep a close eye out for additional compliance frameworks arriving in the near future.
Ultimately, your company may determine that you need to hire a regulatory consultant with specialized knowledge of EU regulations, but as a designer or engineer who wants to stay ahead of this, you should be thinking about how to comply with the AI Act now and how to stay informed as new guidance is released. Figure out what risk level your product may fall under and start preparing to properly manage that risk.
Whether you are about to begin a project or are in progress, take an inventory of every AI feature in your product, including those that may have shipped with a supplier’s hardware or software. This will help you create processes to properly document and verify your product once more regulations are released. If you rely on LLMs, agents, or other third-party AI services to provide functionality, be sure to get compliance documentation from your providers and consult with regulatory experts to figure out what obligations your company may have regarding disclosure and risk management.
Finally, when your product is nearing completion and your marketing team is preparing their materials, make sure to have regulatory experts read all of your marketing copy and your documentation. The current guidance states that a system’s intended purpose is judged from instructions, promotional materials, and technical documentation taken as a whole. Branding alone cannot override the intended purpose shown by your documentation and promotional materials.
One additional practical point: when a manufacturer incorporates third-party AI, its role as the provider, importer, or deployer can determine who is responsible for compliance.
These preparatory steps can result in a smoother launch in the EU even though not all of the final standards of the AI Act have been released. These steps will also make it easier for regulatory experts to interpret where your product falls under the upcoming regulations.
Glossary of Technical Terms
Conformity assessment — Process for demonstrating that a regulated product or AI system satisfies applicable requirements.
Deployers — Organizations or people using AI systems under their authority, except for personal nonprofessional use.
General-purpose Artificial Intelligence (GPAI) — AI models capable of competently performing a wide range of distinct tasks.
Harmonized standards — European standards that can help demonstrate conformity with legal requirements when appropriately referenced.
High-risk AI — AI systems meeting the criteria in Article 6 and the relevant annexes of the Act.
Intended purpose — The intended use specified by the provider, including relevant information in documentation and promotional materials.
Machine-readable marking — Technical marking of AI-generated content that supports automated detection.
Provider — Entity that develops or has an AI system developed and places it on the market or puts it into service under its name.
Glossary of Proper Nouns
AI Omnibus / Digital Omnibus on AI — Regulation (EU) 2026/1744 amending the AI Act and related laws.
CEN and CENELEC — European standardization organizations.
European Commission — Executive institution of the European Union.
EU Artificial Intelligence Act — Regulation (EU) 2024/1689, as amended.
Mistral AI — Developer of AI models.
Meta — Technology company and developer of AI models.
OpenAI — AI research and deployment company.
References
European Commission: Article 50 transparency
European Commission: AI Act timetable
European Commission: Enforcement start dates
European Commission: Article 25 value-chain responsibilities