The European Union has approved regulations for artificial intelligence (AI), becoming the first framework to govern AI. The regulations aim to make AI more “human-centric” and place restrictions on high-risk AI systems. Some AI applications will be banned, while others will require labels. The AI Act is expected to become law in May, with complete regulations in effect by mid-2026. Violations of the AI Act can result in fines of up to $38 million or 7% of a company’s global revenue.
FCC Chairwoman Jessica Rosenworcel has proposed a regulation requiring broadcasters to disclose the use of artificial intelligence (AI) in political advertisements. The proposal seeks consumer input on disclosure methods and the definition of AI-generated content. It mandates on-air and written disclosure in broadcasters’ political files, applies to candidate and issue ads, and includes cable operators, satellite TV, and radio providers.
The UK’s AI Safety Institute (AISI) has found that major LLMs (Language Models) are highly vulnerable to basic jailbreaks, with some models generating harmful outputs without any attempts to produce them. The AISI plans to test other AI models further and develop more evaluations and metrics to address these safety concerns.
Why do we care?
The Europeans continue to lead on regulation, and considering that all the companies offering AI models do business there, EU laws apply. So, that’s why we care.
AI safety is taking a beating – I wanted to mention the recent disbanding of OpenAI’s super alignment team. Semafor even covered the mood of a recent conference – reflecting the pressure, and a sentiment in this quote “When humans invented computers, or for that matter the printing press, very few safety issues were worked out beforehand, nor could they have been.”
Sure… but there’s work to be done here.