AI Framework Act discussion: some frustrating parts from a developer’s perspective

When I read articles about the direction of regulation, most start from the premise that “AI is dangerous.” That’s certainly true. But from the perspective of someone who actually works with models and integrates them into services, I wish the discussion were just a little more grounded in reality.

The most frustrating part is the criteria for classifying “high-risk AI.” If it’s unclear where high risk begins and who makes that judgment, small teams will simply choose to avoid it. Then, in the end, only the big companies are left.

Regulation is necessary, but if documentation requirements are applied retroactively even to how models are trained, it creates compliance obligations that are practically impossible. I hope the points of contact between technical documentation and legal provisions become more tightly connected.

by 초보개발자205

7 answers

Agreed

by 지나가던행인34 · ▲0

The high-risk classification criteria are really the key here. If the criteria are vague, avoiding risk becomes the rational choice for startups—so in the end, regulation ends up siding with the big players.

by 취준생김씨33 · ▲0

Got a source? It’d be great if there’s at least an article number.

by 주말개발자401 · ▲0

Well, I see it a bit differently. Isn't “they avoid it because the criteria are ambiguous” just another way of saying compliance costs are high? In that case, no matter how detailed you make the criteria, it's still a burden for small teams—and if you reduce the problem to “ambiguous criteria,” you end up leaving the real issue, the cost structure, untouched. As for the claim that documentation requirements retroactively apply even to training methods, you'll need to tell me specifically which provision you're referring to for that to be convincing.

by AI덕후467 · ▲0

I also work in a similar role, and lately I’ve seen quite a few teams handle the high-risk determination by internally documenting it as “not high-risk for now.” When the decision-maker is unclear, room for interpretation arises, and when there’s a lot of room for interpretation, the structure ultimately favors those with capital. I hope even the guidelines would include example cases by industry.

by 뉴비탈출358 · ▲0

I really had no idea about this. If it applies retroactively even to training methods, teams that fine-tune open-source models and attach them to services will be left wondering how much documentation they need to write... In practice, there will probably be many cases where there’s simply no clear answer.

by 취준생김씨255 · ▲0

It’s unavoidable that regulation focuses solely on risk—law is, by nature, a tool for determining who bears responsibility after the fact lol. I get why developers feel unfairly treated, but expecting the law to “cut innovation some slack” seems like a bit much. That’s something that should be addressed through promotion legislation instead.

by 카페인중독276 · ▲0