Back to Markets
Stocks● Neutral
Europe and China Offer Two CLM Models. The US Offers None.
The EU's AI Act classifies by risk; China requires approval; the US has none. CLM disinformation surged in 2024 elections, deepening power asymmetries.
Continue with
Roughly three approaches to CLM governance took shape by 2024. The European Union enacted the AI Act that year, classifying systems by risk level. High-risk uses such as biometric recognition and critical infrastructure face strict rules. Transparency is mandatory: users must know when they interact with a CLM. In practice, significant gaps exist. Enforcement capacity is limited. Companies lobby aggressively. Technology moves faster than regulation.
China took a different path. Every CLM system must pass government approval before deployment. Outputs must align with 'socialist core values.' In practice, specific topics are blocked. Other queries go unanswered. The control is openly declared. Transparency here means censorship, not freedom.
The United States has no comprehensive federal regulation. A handful of states passed their own laws. Companies make voluntary commitments. The market, the argument goes, will filter out bad systems. The problem: markets optimize for short-term returns. Polarization and disinformation are long-term social effects that markets ignore.
From EU Risk Rules to Chinese Censorship
None of these models fully addresses the technology's trajectory. The EU framework requires human oversight but depends on enforcement that is understaffed. China's model guarantees alignment with state interests but blocks legitimate uses. The US approach assumes self-regulation works, even as lobbying shapes the rules.
2024 offered a stress test. Over 60 countries held national elections. CLM-driven disinformation appeared in nearly all of them. Auto-generated social media posts and interviews that looked real but never happened. Deepfakes showing politicians saying things they never said. What made these dangerous was ambiguity: content mixed fact with distortion, targeted emotional responses, manipulated context. Detection tools struggle to keep up. The economics of disinformation have changed radically. Production cost is near zero. Scale is enormous.