OpenAIのSolにはAI安全ガードレールが存在しない――その理由は誰も知らない
政府の評価プロセスが不透明なまま、OpenAIのSolとAnthropicのFableは一般公開が承認された。この不透明性は、政治的影響の存在や、フロンティアAIモデルに対する標準化された安全プロトコルの欠如について疑問を投げかけている。
openai and anthropic have secured government clearance for their frontier models, sol and fable, despite a complete lack of public clarity on the specific safety thresholds or evaluation methodologies required. sam altman cited conversations with commerce secretary howard lutnick, but openai refused to detail the process, relying instead on third-party audits from groups like the uk aisi. the trump administration's executive order explicitly ruled out an fda for ai, leaving oversight fragmented across six cabinet agencies. this ad hoc system, compounded by reports of openai's political donations, creates an environment where access to powerful models is determined by unclear backchannels rather than established technical safety standards.