The AI Act takes one step closer

Political agreement on the AI Act means it is likely to come into force and protect us all from Skynet or strangle us in red tape (depending on who you talk to) in 2025 or 2026. There has been plenty of coverage of what is in it and what the purpose is behind it but this article in the FT highlights the real concern, this time voiced by Emmanuel Macron, that it may stifle innovation in the EU which, let's face it, isn't the AI powerhouse the US and China are shaping up to be anyway. 

Other concerns are likely to be:

1. As with GDPR, how practical will it be to comply with and enforce?
2. Will AI's exponential rise make any regulations obsolete and unworkable before they come into force?
3. Regulatory frameworks are put in place to guide good actors and rein in bad ones but bad actors tend to ignore regulations so how much practical protection will regulatory oversight give? 
4. AI developed for internal research or open source purposes won't be subject to the same restrictions. Could this just be a way of getting legally suspect applications on the market by the back door?

As someone who spent a lot of time advising clients on GDPR and helping with damage limitation when they got it wrong, I look forward to a similar experience with the AI Act. Sort of.