Japan's AI plan shows how soft law acquires teeth

Soft law does not usually become hard law on purpose. It becomes hard law through a built-in trigger: a clause that says the system will be reviewed until the system is not soft anymore.

Its December 2025 AI Basic Plan was directional, a national statement of intent without penalties. The cabinet's first revision on 15 July 2026 did not add rules. It added a review trigger: language instructing the government to "proactively and continuously review the system, including the AI law, to deal with risks more effectively," as Asia News Network's reporting on the revised AI Basic Plan records. The mechanism is ratchet-shaped. A plan names a goal. A review clause obliges the state to revisit it on a schedule. An expanded evaluator, in Japan's case the Japan AI Safety Institute (J-AISI), builds the technical capacity to grade frontier models. A future statute supplies the penalties. None of the steps, taken alone, looks like a crackdown. Together they are the legislative path by which soft law graduates into enforcement.

The forcing function this time is cyber, not safety theater. The revised plan names AI-enabled cyberattack risk as the most acute threat, and asks for early access to frontier models under that pretext. Watch the ratchet elsewhere: the EU AI Act, the UK AI Bill, and US state AI bills all carry the same review clause, and most of them are also building the evaluation bodies that make those clauses mean something. The plan does not change the law. The plan changes how soon the law can change.