Companies utilizing AI systems deployed before the AI Act must navigate new compliance requirements triggered by significant modifications to their existing software, design, or original intended purpose. <\/p>\n
Organizations that implemented AI systems before the EU AI Act\u2019s enforcement are not always required to overhaul them. The legislation provides a transition period for existing high-risk systems, though this protection is conditional. Substantial design changes or modifications to the system\u2019s purpose may force businesses to comply with new standards, including conformity assessments, documentation, and risk management.<\/p>\n
For enterprises, determining when a system was launched is no longer sufficient. It is equally critical to track whether the software has been expanded or repurposed. The AI Omnibus further clarifies these rules for systems already on the market before the high-risk requirements become mandatory.<\/p>\n
The AI Act identifies a “substantial modification” as a change made after deployment that was not covered in the original assessment and affects the system’s compliance or intended use. Not every update triggers these obligations; routine bug fixes or security patches are generally exempt, whereas expanding the scope of analyzed data or adding decision-making functions may initiate a full regulatory review.<\/p>\n
Modifying a tool’s purpose is particularly sensitive. For example, a system designed for document sorting cannot be repurposed to evaluate job candidates or credit applicants without meeting strict new requirements.<\/p>\n
The 2026 AI Omnibus specifies that transition periods apply to the type and model of the system rather than individual units. If one unit of a high-risk model was legally deployed before the deadline, subsequent units of the same type may benefit from the transition period, provided the design remains unchanged.<\/p>\n
However, substantial design alterations override these protections, triggering the obligation to comply fully with AI Act requirements for high-risk systems, including mandatory conformity assessments.<\/p>\n
The European Union has adjusted the implementation schedule for high-risk AI systems. For standalone systems covered under Annex III\u2014including those used in employment, education, critical infrastructure, and law enforcement\u2014the compliance date is now December 2, 2027. Systems integrated into regulated products or safety components have until August 2, 2028.<\/p>\n
Despite these extensions, businesses must remain vigilant. Certain prohibitions regarding unacceptable AI practices and requirements for general-purpose AI models are already in effect as of February and August 2025, respectively.<\/p>\n
Companies should conduct an audit of all active AI systems, moving beyond license lists to evaluate deployment dates and modification histories. A lack of documentation regarding how a system evolved over several years poses a significant legal risk, as it may be impossible to prove that a tool qualifies for the transition period.<\/p>\n
The safest approach is to maintain a rigorous history of changes, including vendor agreements, version logs, and performance data. Decisions regarding whether a modification is “substantial” should involve a cross-functional team including IT, legal, and operational leadership.<\/p>\n","protected":false},"excerpt":{"rendered":"
Companies utilizing AI systems deployed before the AI Act must navigate new compliance requirements triggered by significant modifications to their existing software, design, or original intended purpose. Transition Period for Legacy AI Systems Organizations that implemented AI systems before the EU AI Act\u2019s enforcement are not always required to overhaul them. The legislation provides a […]<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[10],"tags":[],"class_list":["post-10703","post","type-post","status-publish","format-standard","hentry","category-tech"],"_links":{"self":[{"href":"https:\/\/bizonews.com\/pl\/wp-json\/wp\/v2\/posts\/10703","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/bizonews.com\/pl\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/bizonews.com\/pl\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/bizonews.com\/pl\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/bizonews.com\/pl\/wp-json\/wp\/v2\/comments?post=10703"}],"version-history":[{"count":0,"href":"https:\/\/bizonews.com\/pl\/wp-json\/wp\/v2\/posts\/10703\/revisions"}],"wp:attachment":[{"href":"https:\/\/bizonews.com\/pl\/wp-json\/wp\/v2\/media?parent=10703"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/bizonews.com\/pl\/wp-json\/wp\/v2\/categories?post=10703"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/bizonews.com\/pl\/wp-json\/wp\/v2\/tags?post=10703"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}