International Relations25 July 20264 min read
Nobel laureates and religious leaders sign Rome Declaration seeking AI curbs on nuclear launches
The Rome Declaration for an Unarmed and Disarming Peace, signed in July 2026 by Nobel laureates, AI scientists, religious leaders and other global figures, calls for an international treaty that bars autonomous AI systems from accessing or deciding on nuclear weapon launches. It also asks for meaningful human control over autonomous weapons. The signatories warn of a "convergent peril" arising from the growing practice of delegating moral and ethical decisions to AI systems.
The Declaration is inspired by Pope Leo XIV's encyclical Magnifica Humanitas, described as a "clarion call" to shield human dignity from technological overreach. It is a normative "nudge" and not a binding legal framework, because there is no global consensus on regulating autonomous AI in warfare.
What the Declaration asks for
Its principles cover disarming the next arms race, responsible development, responsible use, responsible governance, responsible leadership and nuclear disarmament. It sets out five mandates:
- Meaningful human control: an absolute prohibition on automated or AI-driven systems making the final decision to deploy nuclear weapons.
- A "digital commons" model: democratising data so independent experts can better analyse the threats and unintended consequences of ungoverned AI.
- Responsible development: AI developers must publish their ethical frameworks, and fully autonomous, self-improving systems that humans cannot audit or abruptly halt are forbidden.
- Vulnerability audits: nuclear-armed states must review their command-and-control networks to shield them from AI-driven cyber-tampering.
- Time-bound, verifiable disarmament: renewed good-faith negotiations to eliminate nuclear arsenals completely, irreversibly and verifiably.
Why the signatories see urgency
Deterrence has historically relied on the "safety valve" of human hesitation, as in the 1962 Cuban Missile Crisis and the 1983 Soviet false alarm. AI battle management systems compress decision times from hours to seconds, which removes that buffer.
Deep neural networks are also susceptible to "hallucinations" and "data poisoning". In a crisis, an AI might read a corrupted image or a routine network ping as an incoming missile and trigger retaliation before humans can intervene.
Arms-control treaties such as Open Skies and New START are failing, which gives states an incentive to automate their defences to match the perceived speed of adversaries. In contested settings such as the Iranian nuclear programme, where diplomatic buffers have broken down and states increasingly use AI to monitor facilities, an automated defence protocol could misread a digital intrusion or physical strike and respond with a disproportionate missile launch.
The wider concern about outsourcing ethics
A convergent peril describes several risks that interact and amplify one another: algorithmic bias, loss of human accountability, hyper-automation and ethical abdication.
The concerns include "moral deskilling", the loss of the human capacity for ethical reasoning. There is also a value-alignment gap: AI models tend towards utilitarianism, while human justice rests heavily on deontological ethics, so an AI might justify sacrificing a minority's rights for a majority outcome. Models trained on historical data can carry over prejudices of caste, race and gender under the guise of objective mathematics. Deep learning "black boxes" leave it unclear who is culpable for a wrong decision, and AI cannot apply mercy, equity or contextual nuance.
Where existing frameworks stand
The EU AI Act, 2024 takes a risk-based approach, banning unacceptable-risk AI such as social scoring and imposing strict obligations on high-risk systems. The Bletchley Park Declaration (2023), signed by 28 countries including India, recognises the risks of frontier AI and promotes cooperation on AI safety. The UNESCO Recommendation on the Ethics of AI (2021) promotes human dignity, transparency, accountability, human rights and sustainability. India is a founding member of the Global Partnership on AI.
In India, NITI Aayog's National Strategy for AI (2018) set out the "AI for All" vision. The IndiaAI Mission (2024) includes a "Safe and Trusted AI" pillar. The Ministry of Electronics and Information Technology's India AI Governance Guidelines follow a principle-based, techno-legal framework anchored in the Seven Sutras.
Steps proposed to close the gap
The measures put forward are mandatory human oversight for decisions touching life, liberty or dignity, explainable AI that third-party regulators can audit, and ethical impact assessments before deployment, modelled on environmental impact assessments. They also include rights-respecting design and a swift, binding UN convention banning Lethal Autonomous Weapons Systems (LAWS), where the decision to kill is outsourced to machines.