EU AI Act Deadlines Shift, But Immediate Compliance for Core Ethical AI Practices Remains Critical

The European Union's AI Act continues its phased implementation, with recent adjustments to deadlines for certain categories of AI systems. Specifically, the Digital Omnibus on AI, which became effective on July 27, 2026, has pushed back the application dates for stand-alone high-risk AI systems (like those used in hiring or credit scoring) to December 2, 2027. Product-embedded AI, such as that found in machinery and medical devices, now faces an August 2, 2028, deadline. However, it is crucial to note that several core ethical provisions of the Act have been in force since February 2025, including outright bans on AI practices deemed to pose an unacceptable risk, such as manipulative or deceptive techniques, exploitation of vulnerabilities, social scoring, and predictive policing based solely on profiling. General-purpose AI rules have also been active since August 2025, and requirements for chatbot disclosure and deepfake labeling were introduced in August 2026. This staggered implementation has significant implications for practitioners. While the extended deadlines for high-risk systems might suggest a reprieve, the immediate applicability of fundamental ethical guidelines means that organizations must already be actively assessing and adjusting their AI deployments. The Act covers providers, deployers, importers, and distributors, extending its reach even to non-EU startups whose AI output is used within the EU. This means that even using a tool like ChatGPT for paying clients can classify an organization as a 'deployer' with associated transparency and AI literacy duties. The emphasis on human oversight for high-risk AI systems, for instance, is not merely a compliance checkbox but a design principle that fosters trust and defensibility. This development fits into a broader trend of increasing global regulatory scrutiny on AI ethics. As AI systems become more pervasive, governments and regulatory bodies worldwide are grappling with how to ensure responsible development and deployment. The EU AI Act, with its comprehensive framework, is a leading example, influencing discussions and legislative efforts in other regions. The ongoing debate around AI safety bills and the authority to halt dangerous AI systems further underscores this trend, highlighting the growing recognition that self-regulation alone is insufficient for mitigating the potential harms of advanced AI. The focus is shifting from simply technological capability to the societal impact and ethical implications of AI. In practice, this means that organizations should not view AI Act compliance as a distant future problem. Instead, they should immediately undertake a thorough inventory of all AI tools they build, buy, or use, assigning roles and risk tiers to each. Any features that could violate the Article 5 bans (e.g., manipulative AI) must be removed. Furthermore, practitioners need to implement clear labeling for bots and AI-generated content and provide comprehensive training for their teams on AI literacy and ethical use. Requesting detailed documentation from AI vendors is also critical. For any high-risk systems, even those with extended deadlines, starting a technical file and exploring national sandboxes for compliance testing can provide a significant advantage. Ignoring these immediate requirements not only risks substantial fines (up to €35 million or 7% of global turnover) but also jeopardizes public trust and competitive standing in a market increasingly sensitive to ethical AI practices.