Dial E for Ethical Enforcement: institutional VETO power as a governance primitive

Researchers currently lack meaningful power to stop harmful uses of their work: The paper argues that existing safeguards like model cards and responsible AI statements are merely "reputational signals" without actual decision-making authority, meaning researchers cannot formally refuse to allow their AI models to be weaponized or misused even when they identify serious risks.

Institutional veto power—the formal authority to halt deployment of AI systems—should be treated as a fundamental governance tool: Drawing parallels to how nuclear nonproliferation and biomedical ethics established veto mechanisms, the authors propose that AI governance needs similar enforceable checkpoints where designated parties can stop a model's use or distribution when credible weaponization risks emerge.

Current governance structures fail because they lack enforcement mechanisms: The paper identifies that "compliance without enforceable constraints fails," meaning that voluntary commitments to responsible AI are insufficient without actual institutional power to prevent harmful outcomes.

Vulnerable communities must lead the design of these safeguards: The authors emphasize that "communities most vulnerable to military uses must lead governance design," suggesting that those most at risk from weaponized AI should have primary authority in deciding when and how veto power is exercised.

Converting symbolic commitments into real accountability requires structural change: The paper argues that meaningful "model disarmament" and responsible AI cannot be achieved through statements and policies alone—institutional redesign with enforceable veto authority is a prerequisite for converting symbolic safeguards into actual responsibility.