The 𝗘𝗨 𝗔𝗜 𝗔𝗰𝘁 is no longer something to “prepare for someday.” Key provisions have already started applying, including the transparency rules from 2 𝗔𝘂𝗴𝘂𝘀𝘁 𝟮𝟬𝟮𝟲.
And one point is especially important for 𝘀𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀:
𝗬𝗼𝘂 𝗱𝗼 𝗻𝗼𝘁 𝗻𝗲𝗲𝗱 𝘁𝗼 𝘁𝗿𝗮𝗶𝗻 𝗼𝗿 𝗯𝘂𝗶𝗹𝗱 𝗮𝗻 𝗔𝗜 𝗺𝗼𝗱𝗲𝗹 𝘁𝗼 𝗯𝗲 𝗮𝗳𝗳𝗲𝗰𝘁𝗲𝗱 𝗯𝘆 𝘁𝗵𝗲 𝗔𝗜 𝗔𝗰𝘁.
𝗜𝗳 𝘆𝗼𝘂𝗿 𝗮𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝗰𝗮𝗹𝗹𝘀 𝗮𝗻 𝗔𝗜 𝗔𝗣𝗜, 𝗲𝗺𝗯𝗲𝗱𝘀 𝗮𝗻 𝗟𝗟𝗠, 𝗿𝗮𝗻𝗸𝘀 𝗰𝗮𝗻𝗱𝗶𝗱𝗮𝘁𝗲𝘀, 𝗽𝗿𝗼𝗰𝗲𝘀𝘀𝗲𝘀 𝗽𝗲𝗿𝘀𝗼𝗻𝗮𝗹 𝗱𝗮𝘁𝗮, 𝗮𝘀𝘀𝗶𝘀𝘁𝘀 𝘄𝗶𝘁𝗵 𝗳𝗶𝗻𝗮𝗻𝗰𝗶𝗮𝗹 𝗱𝗲𝗰𝗶𝘀𝗶𝗼𝗻𝘀, 𝗿𝘂𝗻𝘀 𝗮 𝗰𝗵𝗮𝘁𝗯𝗼𝘁, 𝗼𝗿 𝘂𝘀𝗲𝘀 𝗔𝗜 𝗮𝘀 𝗽𝗮𝗿𝘁 𝗼𝗳 𝗮 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝘄𝗼𝗿𝗸𝗳𝗹𝗼𝘄, 𝘁𝗵𝗲 𝗶𝗻𝘁𝗲𝗻𝗱𝗲𝗱 𝘂𝘀𝗲 𝗼𝗳 𝘁𝗵𝗲 𝘀𝘆𝘀𝘁𝗲𝗺 𝗺𝗮𝘁𝘁𝗲𝗿𝘀. Using a third-party model does not automatically outsource your responsibilities. The Act explicitly covers both providers and deployers of AI systems used in the EU.
So AI compliance is becoming an engineering concern too: 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲, 𝗱𝗮𝘁𝗮 𝗳𝗹𝗼𝘄𝘀, 𝗹𝗼𝗴𝗴𝗶𝗻𝗴, 𝗵𝘂𝗺𝗮𝗻 𝗼𝘃𝗲𝗿𝘀𝗶𝗴𝗵𝘁, 𝘁𝗿𝗮𝗻𝘀𝗽𝗮𝗿𝗲𝗻𝗰𝘆, 𝘀𝗲𝗰𝘂𝗿𝗶𝘁𝘆, 𝘃𝗲𝗻𝗱𝗼𝗿 𝘀𝗲𝗹𝗲𝗰𝘁𝗶𝗼𝗻, 𝗮𝗻𝗱 𝗱𝗼𝗰𝘂𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻.
I put together this infographic and a few practical examples from a software engineer’s perspective.
AI governance is increasingly becoming part of software engineering, not something added by Legal after the system is built.
