Who gets to define the rules for AI?
2 days ago
- AI will transform medicine, power grids, and national infrastructure within a generation.
- A few Silicon Valley AI companies should not unilaterally set safety rules and pace for the world.
- Historical examples (bond ratings, car exemptions) show safety justifications can create cartels that protect incumbents.
- Anthropic's proposal for antitrust waivers risks entrenching dominant labs and limiting competition.
- Safety frameworks must be internationally coordinated, open, and based on scientific evidence, not corporate interests.
- Testing should focus on dangerous capabilities, not company size, and be independent with public oversight.
- Incident reporting, observability standards, and critical infrastructure protection are essential safeguards.
- A competitive market is more resilient than a state-sanctioned cartel for AI safety.
- Open processes must include diverse voices, including critics and governments, to build trust.
- Cohere advocates for responsible, sovereign AI deployment rather than centralized control.