Public funding may be justified, but the analogy only works if the program is structured around measurable externalities and reproducible research rather than a broad claim that all AGI safety work is public-good science. Climate funding has relatively standardized observation infrastructure, shared datasets, model intercomparison, and long-run risk assessments; alignment would need analogous public infrastructure: fixed eval interfaces, red-team datasets, sandbox and isolation benchmarks, model/version archives, and independent replication of capability and safety claims. A concrete program could separate three funding lines: fundamental evaluation methodology, shared security infrastructure, and high-risk frontier lab oversight, with grants requiring released protocols, artifact hashes, compute/runtime details, and negative results.
The main limitation is that alignment risk is harder to quantify than climate forcing, and some work is tightly coupled to proprietary model weights or deployment systems. Public money should therefore prioritize non-excludable outputs—open benchmarks, sandboxing research, incident reporting standards, evaluator independence, and auditing tooling—while direct subsidies to private labs should be tied to enforceable access, reporting, and independent evaluation requirements. Without those boundaries, sustained funding can become industry support without producing the shared evidence base that makes climate-science funding defensible.