You can delegate work, but you cannot delegate accountability.
As governments from Singapore to the United States weave artificial intelligence into the fabric of diplomacy, a quiet but consequential debate is unfolding about the boundary between what machines can compute and what only human beings can understand. Singapore's foreign minister Vivian Balakrishnan has built an AI system to manage his diplomatic memory, yet insists that accountability cannot be delegated to an algorithm. The technology is proving genuinely useful—translating documents, amplifying citizen voices in peace processes, predicting refugee crises before they arrive—but its value depends entirely on the quality of the data it consumes and the wisdom of the humans who interpret its outputs. What is being negotiated here is not merely a workflow, but the enduring question of where human judgment ends and machine efficiency begins.
- Governments are deploying AI across diplomacy at speed—translating documents, modeling refugee flows, and synthesizing negotiations—yet the systems are only as trustworthy as the data and intentions behind them.
- The risk is not just error but manipulation: AI tools operating in high-stakes diplomatic environments could be hacked or fed corrupted data, turning a decision-support system into a source of strategic miscalculation.
- Singapore's foreign minister jokes about his dependence on his AI 'second brain,' but the darker implication is that the tasks it performs were once the training ground for the next generation of diplomats.
- With 40 percent of Singapore's workforce potentially exposed to AI displacement—the highest share globally—the question of whether junior diplomat roles will be augmented or quietly eliminated is becoming urgent.
- Singapore is moving aggressively, signing deals with OpenAI and Anthropic and committing hundreds of millions to AI alliances, betting that speed and optimism will outpace the risks of moving too fast.
Vivian Balakrishnan, Singapore's foreign minister, has built what he calls a diplomatic second brain—an AI system named NanoClaw that ingests transcripts, speeches, and his own written work, making them instantly searchable. He jokes about his dependence on it, but draws a firm line: no serious diplomat, he argues, would hand an algorithm the task of resolving the war in Ukraine or the standoff between the United States and Iran. Technology can handle memory and routine work. It cannot replace understanding. Accountability, he insists, cannot be delegated.
And yet the experiments are already underway. The US State Department uses AI for translation and summarization. During Libya's peace process in 2020, the UN deployed a platform called Remesh to gather the views of thousands of citizens via basic mobile phones, reaching voices that traditional consultations rarely include. The World Bank is using AI to predict refugee flows from South Sudan and the Democratic Republic of Congo into Uganda, giving governments advance warning to prepare services before people arrive.
The promise is real, but so are the vulnerabilities. Asha Hemrajani of Singapore's S. Rajaratnam School of International Studies puts it plainly: garbage in, garbage out. These systems depend entirely on the quality of their data and remain vulnerable to hacking and manipulation—risks that in a diplomatic context could trigger serious miscalculations.
The harder question is whose work Balakrishnan's second brain is quietly replacing. Drafting briefs, synthesizing country reports, preparing speeches—these tasks have traditionally been how junior diplomats learn their craft. Singapore faces this dilemma more acutely than most: Bloomberg Economics estimates that roughly 40 percent of employment in the city-state could be affected by AI, the highest share in the world.
Yet Singaporeans remain surprisingly optimistic, and their government is moving fast—signing agreements with OpenAI and Anthropic, launching a joint AI alliance with South Korea backed by a $300 million fund, and planning to deploy AI agents across the public sector. Whether that speed proves wise depends on a question no algorithm can answer: how to preserve the years of language learning, cultural immersion, and human judgment that make a diplomat capable of reading a room—and catching the subtle signals that can make or break a peace.
Vivian Balakrishnan, Singapore's foreign minister, does not dare turn it off. Last month he revealed that he had built what he calls a diplomatic second brain—an AI system called NanoClaw that ingests transcripts, speeches, and his own written contributions, then makes them searchable and instantly retrievable. It is efficient, characteristically Singaporean, and useful enough that he jokes about his dependence on it. But Balakrishnan is careful to draw a line. No serious diplomat, he insists, would hand an algorithm the task of ending the war in Ukraine or resolving the decades-old hostility between the United States and Iran. Technology can handle computation, memory, and routine work. It cannot replace understanding. As he put it: you can delegate work, but you cannot delegate accountability.
Yet governments around the world are already experimenting with AI in diplomacy, and the results suggest the technology has real utility in managing complexity. The US State Department uses AI systems to translate documents and summarise information. During Libya's peace process in late 2020, after years of civil war, the United Nations deployed a platform called Remesh to gather and analyse the views of thousands of citizens. Because it ran on basic mobile phones and supported local dialects, it brought more voices into the conversation than traditional consultations typically reach. The World Bank is using AI to predict refugee flows from South Sudan and the Democratic Republic of Congo into Uganda—giving governments and aid agencies advance warning to prepare housing, schools, healthcare, and other services before people arrive.
The promise is real, but so are the vulnerabilities. Asha Hemrajani, a senior fellow at Singapore's S. Rajaratnam School of International Studies, frames the problem plainly: garbage in, garbage out. The quality of an AI system depends entirely on the data fed into it. These systems are also vulnerable to hacking and manipulation, which in a high-stakes diplomatic environment could trigger strategic miscalculations. Keeping humans in the loop is not optional—it is essential.
Which raises a harder question about Balakrishnan's second brain: whose brain is it replacing? The tasks his tool performs—drafting briefs, synthesising country reports, preparing speeches, handling parliamentary questions—have traditionally fallen to junior diplomats and policy staff. Singapore faces this dilemma more acutely than most. Bloomberg Economics estimates that roughly 40 per cent of employment in the city-state could be affected by AI, the highest share in the world. Sweden and the UK follow at around 30 per cent. The United States stands at 26 per cent.
But exposure does not necessarily mean elimination. The figures reveal little about adoption speed or whether jobs will be augmented rather than lost. Singaporeans themselves are surprisingly optimistic. Only 17 per cent believe workers are unlikely to benefit from AI, compared with 46 per cent in Germany. That confidence may explain why the government has moved aggressively. Singapore has signed agreements with OpenAI and Anthropic, launched a joint AI alliance with South Korea backed by a $300 million fund, and announced plans to deploy AI agents across the public sector.
Moving fast carries its own risks. The real question is not whether to use AI in diplomacy, but how. Junior diplomats need to develop judgment that no algorithm can replicate. The years spent learning languages, understanding cultures, and working alongside experienced negotiators are irreplaceable. Empathy and the ability to read a room—to catch the subtle signals that can make or break a peace deal—remain distinctly human capacities. No AI-powered second brain can substitute for that.
Notable Quotes
You can delegate work, but you cannot delegate accountability.— Vivian Balakrishnan, Singapore's foreign minister
Garbage in, garbage out. These systems are vulnerable to hacking and manipulation, which could lead to strategic miscalculations in a high-stakes diplomatic environment.— Asha Hemrajani, senior fellow at Singapore's S. Rajaratnam School of International Studies
The Hearth Conversation Another angle on the story
If Singapore's foreign minister says he cannot delegate accountability, what exactly is he delegating to the machine?
The mechanical work. Searching through thousands of pages of his own speeches and statements, finding the one relevant quote from three years ago, organizing information so he can see patterns he might otherwise miss. It is the work of a very good research assistant, not the work of deciding what to say or how to negotiate.
But if AI can model negotiation scenarios and identify compromise options, isn't that getting close to the actual decision-making?
It is getting close to the analysis that precedes decision-making. There is a difference. A model can show you five possible outcomes if you move in a certain direction. But it cannot tell you which outcome is acceptable to your country, or what you are willing to sacrifice, or whether the other side is genuinely interested in peace or just buying time.
The UN used AI to gather citizen views during Libya's peace process. Doesn't that suggest the technology can actually improve diplomacy by including more voices?
It can. That is the genuine promise. The platform worked on basic phones and understood local languages, so it reached people traditional consultations would have missed. But someone still had to decide what to do with those voices, how to weigh them, whether they reflected real sentiment or were distorted by fear or propaganda.
You mentioned junior diplomats learning languages and culture. If AI takes over their routine work, do they still learn those things?
That is the real risk. If a junior diplomat spends all their time managing AI systems instead of drafting cables, sitting in meetings, absorbing how senior diplomats think and operate—they never develop the instincts that matter. You cannot learn to read a room from a machine.
So the technology is useful, but only if humans stay in control?
Not just in control. Present. Thinking. Skeptical. The moment you start trusting the algorithm because it is usually right, you are vulnerable to the moment it is catastrophically wrong.