Getty
The European Union made history Wednesday, passing the world’s first comprehensive law regulating artificial intelligence. The AI Act is expected to officially become law by May or June, pending formal approval from EU member countries, with provisions starting to take effect six months after the rules enter the lawbooks.
Rules for general-purpose AI systems like ChatGPT will start applying a year after the law takes effect. The complete set of regulations is expected to be in force by mid-2026.
The EU has taken a risk-based approach to its AI regulation, ranking the use of artificial intelligence according to the potential damage it could do. The law bans AI systems that carry “unacceptable risk,” including using biometric data to detect a person’s ethnicity or sexual orientation. High-risk applications, including the use of AI in hiring or law enforcement, will be more tightly regulated, with developers required to show that their models are safe and transparent and adhere to privacy regulations.
For lower-risk AI tools, there will be little to no regulation but developers will still be required to label AI-generated deepfake pictures, video, or audio of existing people, places, or events as artificially manipulated. The law applies to models operating in the EU and any firm that violates the rules risks a fine of up to 7 percent of its annual global profits.
When it comes to enforcement, each EU country will set up its own AI watchdog, and citizens will be able to file complaints if they think their rights have been violated. Brussels will create a stand-alone AI Office tasked with enforcing and supervising the law for general-purpose AI systems.
There will be extra scrutiny for the biggest and most powerful AI models that the EU judges to pose “systemic risks,” which include OpenAI’s GPT4 and Google’s Gemini. Companies providing these systems will have to assess and mitigate the risks, put cybersecurity measures in place, report any serious incidents as a result of their systems and disclose how much energy their models use.
All general-purpose AI systems will have to draw up a policy showing that the content used for training their models respects European copyright law.
As with other digital regulations — like last year’s Digital Services Act, which targeted abuse on social media or the Digital Markets Act, which went into effect March 7 and has the goal of combating market dominance by so-called digital “gatekeepers” — the EU’s AI act aims to be the global default legislation. Similar laws are on the way in countries from Brazil to Japan.
In October, U.S. President Joe Biden signed a sweeping executive order on AI that is expected to be backed up by legislation and global agreements.
