EU Parliament eases AI Act rules and bans deepfake undress apps

The EU’s latest tweak to its AI Act is a two‑sided signal. On one side, Parliament is visibly responding to public outrage over AI‑generated sexual abuse, especially the Grok deepfake scandal, by explicitly banning “nudification” apps that undress people without consent. That reflects a hardening view that some AI use cases are simply incompatible with fundamental rights, and that platforms and model providers will be held responsible for how their tools can be abused.

On the other side, the package quietly defers and simplifies several compliance obligations for high‑risk AI systems, particularly where requirements overlap with existing sectoral rules for machinery or medical devices. By pushing some deadlines out to 2027–2028 and clarifying that not every AI‑enabled feature counts as a high‑risk safety component, lawmakers are trying to ease industry’s complaints that the original AI Act was too rigid. For frontier labs and large deployers, this combination means a slightly more forgiving compliance runway, but with sharper red lines around specific harms. It’s a reminder that the regulatory environment for advanced models won’t be purely restrictive or enabling—it will selectively criminalize certain applications while giving breathing room to others.