Adopting AI in public services: make your pilot assessable

This analysis first appeared as the weekly Praxikon newsletter.

<h2>Brussels focuses on AI in public services</h2> <p>On 9 September, the European Commission announced an <a href="https://digital-strategy.ec.europa.eu/en/events/apply-ai-webinar-sectoral-deep-dive-public-sector">Apply AI webinar on the public sector</a> for 30 September. Its focus is practical adoption, interoperability and European open-source solutions. The agenda also includes the AI Toolbox for public services, the PAIR pathway and GenAI pilots. This is an implementation initiative, not a new AI Act deadline.</p> <h2>What does this mean for you?</h2> <p>Our practical interpretation: decide upfront what you will assess in an AI pilot. An application that only drafts a response needs different checks from one that ranks applications or prepares a decision. The product name alone tells your reviewers too little.</p> <p>This also matters for advisers and training providers supporting public organisations. Let participants practise with a recognisable workflow: what information does the system receive, what does it produce, who checks the result and what happens when it goes wrong? This connects training directly to implementation.</p> <h2>Which timeline applies?</h2> <p>The <a href="https://eur-lex.europa.eu/eli/reg/2026/1744/oj">Digital Omnibus on AI</a> has been in force since 27 July 2026. For high-risk systems under Article 6(2) and Annex III, the deferred obligations in Chapter III, Sections 1 to 3, apply from 2 December 2027. For Article 6(1) and Annex I, the date is 2 August 2028.</p> <p>That extension does not cover everything. The <a href="https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act">official implementation timeline</a> confirms that Article 50 on transparency has applied since 2 August 2026. The transition until 2 December 2026 covers only Article 50(2) for the relevant systems already placed on the market before 2 August 2026. Do not treat it as a general extension for chatbot disclosures or other transparency obligations.</p> <h2>Action for this week</h2> <p>Take one planned pilot and spend 30 minutes reviewing it with the process owner, IT and a legal reviewer. Write down what the application may and may not do. Then choose one realistic error, such as an incorrect summary of an application. Test whether the responsible employee spots the error, can correct it and can stop the process if necessary.</p> <p>Record the outcome with a test example, an owner and an outstanding improvement action. This is our practical advice for preparation, not a legally prescribed documentation format. It does provide a concrete basis for determining which rules relate to this application and your role.</p> <p>Use <a href="https://www.praxikon.com/en">Praxikon to explore your AI Act question with sources</a>. Include the specific intended use in your question.</p> <p>You receive this weekly AI Act update because you signed up with Praxikon, formerly Praxikon, or used one of our tools or templates. You can unsubscribe with one click at the bottom.</p> <p>Until next Tuesday.<br/>Zahed Ashkara<br/>Praxikon</p>