AI Safety Employees Could Unionize to Enforce Safety Commitments
What if researchers could collectively threaten to quit when companies race to AGI?
Deep Dive
Boyd Kane argues that AI safety employees face a repeated problem: companies talk safety to hire talent, then renege. Unions or union-like structures could let researchers collectively threaten to quit over red lines, but unions have legal limits—they can't veto business strategy, exclude supervisors, and visa risks weaken threats. A more realistic alternative: public, pre-signed agreements among employees to resign if certain red lines are crossed, making the threat credible and predictable to employers.
Key Points
- Unions under NLRA cannot veto business strategy, only wages and hours
- Supervisors and visa-holders face barriers to unionizing
- Public resignation pacts offer a credible threat without legal hurdles
Why It Matters
A new tool for AI researchers to hold companies accountable for safety promises, potentially slowing reckless AI development.