To make AI safe, we must develop it as fast as possible without safeguards (alignmentalignment.ai)

An opinion piece by an AI company leader argues that the safest path to artificial general intelligence (AGI) is the opposite of cautious development: race to build it as quickly as possible and strip away safeguards. The author—who previously supported a six‑month moratorium—uses the Manhattan Project and nuclear deterrence as a model, claiming that rapid, competitive development and proliferation of powerful systems will prevent a single bad actor from gaining the ability to annihilate humanity. The piece explicitly rejects slower, society‑wide deliberation and regulatory safeguards, endorsing a “release now, regulate later” or “stable door” approach and suggesting multiple AGIs could hold each other in check. This argument matters because it directly counters mainstream AI safety and governance priorities and could influence investors, startups, and researchers tempted to prioritize speed over robustness. Technically, the proposal foregrounds build‑first strategies for superintelligence rather than staged safety engineering, oversight, or verification—increasing incentives to cut corners on testing, alignment, adversarial robustness, and access controls. For the AI/ML community and policymakers the implications are stark: a rhetoric that normalizes arms‑race dynamics raises the probability of catastrophic failures, accidental misuse, and geopolitical escalation, while also exposing the limits and moral risks of historical analogies when applied to existential AI risks.