Regulators cannot rely on intent alone. Companies often wrap defensive lobbying in the language of 'existential risk' to pull focus away from the data scraping, copyright theft, and algorithmic bias happening right now. To tell them apart, look at the technical substance. Legitimate alignment research produces open-source benchmarks, peer-reviewed methodology, and reproducible results. It focuses on the mechanics of control and predictability.
Lobbying, by contrast, often pushes for high barriers to entry under the guise of safety. If a firm argues that only massive, centralized players should be allowed to train large models because they are the only ones who can 'safely' manage them, they are likely engaging in regulatory capture. They use long-term fears to build a moat around their current market share. They want to make it too expensive for small competitors to operate.
Effective oversight requires a two-track approach. One track must address immediate harms like misinformation and labor exploitation through strict, enforceable standards. The second track should support fundamental safety research through public funding rather than relying solely on industry-led initiatives. When safety proposals prioritize market dominance over transparent technical verification, regulators should treat them as political maneuvers, not scientific imperatives.
