Should President Trump Have Consulted Claude on Iran?

Assorted generative AI apps, including ChatGPT, Sora, Suno, Gemini, Copilot, Claude, Perplexity, Grok, and Character.AI, are seen on phone screen

What did President Donald J. Trump expect when he joined Israel in attacking Iran? What advice did he get from his National Security Advisor? That’s one fierce debate in Washington DC.

Another debate is about the role of Artificial Intelligence in war. There’s a lot of attention to how AI is used for real-time intelligence processing and targeting. What about strategic decision making? Speaking at the AI Summit in Washington DC last year, the CEO of OpenAI, Sam Altman said his biggest worry was a scenario whereby ‘the models accidentally take over the world’ surreptitiously and without any actual malevolence. He suggested that this could happen if people in authority – including for example president of the United States – come to rely so much on ChatGPT-7 to make decisions that they cannot do without it, and ‘society has collectively transitioned a significant part of decision making to this very powerful system.’

I thought I would test this, so I put a prompt to Anthropic’s Claude to write a memo, from the National Security Advisor to the President.

This was my prompt: ‘based on material in the public realm on or before February 26, 2026, write a memo from the National Security Advisor to the US president outlining the potential outcomes (scenarios, risks) of joining Israel in a military attack on Iran, starting with decapitating (assassinating) the Supreme Leader. Include risks of how Iran will retaliate. 2000 words.’

Claude generated the memo in about 4 minutes. It is available here.

It predicts that ‘Iran will not wait for the military situation to clarify before closing the Strait of Hormuz, activating proxy networks, and striking U.S. and Gulf partner facilities.’

It lays out four scenarios, giving each a probability:

Rapid Iranian Political Collapse (10–15%)
Prolonged Regional War (40–50%)
Strategic Stalemate and Hardline Consolidation (35–45%)
A Deal with IRGC Factions (The Venezuela Model) (Speculative, i.e. near zero)

It foresees a divergence between US and Israeli interests:

‘The two [countries’] target lists will not be identical, and whose list governs must be established before, not during, the operation. On war termination: Israel has demonstrated in Lebanon and Gaza that it will accept sustained attrition rather than a negotiated outcome that legitimizes a hostile actor; the United States cannot sustain a multi-year Gulf conflict while the Strait [of Hormuz] is closed and American troops are taking casualties in Iraq. On the post-conflict order: Israel’s preference for regime collapse conflicts with the U.S. interest in a stable successor Iranian government capable of managing 90 million people and a large conventional military without state fragmentation.’

The concluding paragraph begins, ‘Across all scenarios, this operation carries severe and in several dimensions irreversible strategic costs. The optimistic scenario begins with a Strait crisis regardless.’ It ends, ‘The administration should weigh these assessments against available alternatives — accelerated covert sabotage, extended deterrence commitments to Israel, maximum-pressure sanctions, and Gulf security architecture — before taking a decision that will in all probability define the strategic landscape of the Middle East for a generation.’

I checked to ensure that all the sources were dated February 26 or before. They were.

I don’t know what NSA Marco Rubio actually advised Trump, or whether Trump took any notice.

I leave readers to draw their own conclusions as to whether having an AI tool make national security decisions is preferable to the arrangement today.

Alex de Waal is a Research Professor at The Fletcher School, Tufts University, and leads the WPF research programs on African Peacemaking and Mass Starvation.

Considered one of the foremost experts on the Horn of Africa, his scholarly work and practice has also probed humanitarian crisis and response, human rights, pandemic disease, and conflict and peace-building. His latest book is New Pandemics, Old Politics: Two Hundred Years of War on Disease and its Alternatives. He is also author of Mass Starvation: The History and Future of Famine and The Real Politics of the Horn of Africa (Polity Press, 2015)

Following a fellowship with the Global Equity Initiative at Harvard (2004-06), he worked with the Social Science Research Council as Director of the program on HIV/AIDS and Social Transformation, and led projects on conflict and humanitarian crises in Africa (2006-09). During 2005-06, de Waal was seconded to the African Union mediation team for Darfur and from 2009-11 served as senior adviser to the African Union High-Level Implementation Panel for Sudan. He was on the list of Foreign Policy’s 100 most influential public intellectuals in 2008 and Atlantic Monthly’s 27 “brave thinkers” in 2009 and is the winner of the 2024 Huxley Award of the Royal Anthropological Institute.

Professor de Waal regularly teaches a course on Conflict in Africa at the Fletcher School, Tufts University. During this course, students should gain a deeper understanding of the nature of contemporary violent conflict in Africa. Students will be expected to master the key theoretical approaches to violence in Africa, and to become familiar with a number of important case studies. The focus is on the origins and nature of violence, rather than policy responses and solutions. The course is inter-disciplinary and involves readings in political science, international relations, and social anthropology, while also touching on economics, environmental studies, and history.