SEOUL, October 23 (AJP) - ChatGPT and Claude vie for headlines and users with faster and smarter upgrades, but in South Korea Gemini suddenly came into focus as an AI sage after President Lee Jae Myung disclosed he had consulted the Google chatbot on a matter of state.
Lee, an avid social media user, wrote earlier this week that he had asked Gemini whether the heads of new prosecutorial and investigative agencies should attend weekly Cabinet meetings.
He shared what he described as Gemini’s conclusion: arguments against their attendance were constitutionally unreasonable and contradicted the president’s authority to supervise government agencies.
The disclosure brought a familiar technology into more sensitive territory.
Seeking an answer from a chatbot is now routine. Citing that answer to defend the exercise of presidential authority raises a different question: where does advice end and authority begin?
Kim Ji-yong, the inaugural head of the Serious Crimes Investigation Agency, attended Tuesday’s Cabinet meeting at Cheong Wa Dae, drawing objections that the appearance could compromise the political neutrality expected of an investigative chief.
Lee argued that investigative bodies should not be treated differently from other government agencies, citing Gemini’s response. A presidential spokesman said Lee had consulted AI to clarify legal ambiguities, rather than to make a judgment or decision.
The distinction lies at the heart of the debate. A chatbot may offer a second opinion, but how much weight should a president give its words — and does quoting them lend a machine the wisdom of Solomon?
A similar question surfaced in Europe last year.
In August 2025, Swedish Prime Minister Ulf Kristersson faced criticism after revealing that he regularly consulted AI chatbots, including OpenAI’s ChatGPT and the French-developed Le Chat, for alternative perspectives on government matters.
“I use it myself quite often. If for nothing else than for a second opinion,” Kristersson told Swedish business newspaper Dagens industri.
His admission drew objections from technology researchers and commentators who questioned whether AI systems could offer meaningful political judgments. His spokesperson said the prime minister did not enter security-sensitive information into the systems.
In South Korea, the debate comes as generative AI becomes an increasingly ordinary part of working life.
The country’s generative AI usage rate surpassed 40 percent and recorded the fastest growth in Microsoft’s AI Economy Institute’s second-quarter 2026 Global AI Diffusion Report.
A recent SpaceAdd survey of 881 office workers found that 95.2 percent used AI for work, with 62.4 percent paying for AI services.
Lawmakers are no exception.
“I use it for searches and research,” said Son Myoung Soo of the ruling Democratic Party.
Some ruling party lawmakers argued that a president seeking to make South Korea an AI powerhouse should be familiar with the technology.
“It is desirable for the president supporting sovereign AI foundation model to use generative AI,” said Kim Hyun, also of the Democratic Party.
Opposition lawmakers acknowledged using the same tools, while questioning the weight a president should attach to their answers.
“I use AI myself, but only as a reference,” said Jung Sung-kook of the People Power Party.
“AI tends to understand the user’s intentions and provide tailored answers. A president must therefore exercise even greater caution when using AI,” Jung said.
“AI should play a supporting role. It must not dominate human insight and judgment,” he added.
Such concerns cross party lines.
“We have reached a point where the scope and extent of AI use should be strictly regulated,” said Min Hong-chul, a Democratic Party lawmaker.
“AI is a tool intended to benefit humanity and enhance human dignity. We must prevent a reversal of roles in which humans become dominated or subjugated by AI and lose what makes them human,” Min said.
Fellow Democratic Party lawmaker Jung Jin-wook said that however sophisticated the technology became, the president and presidential aides must retain responsibility for thinking through decisions.
“As AI becomes more advanced, presidents will grow more dependent on it. At that point, we will need a public debate about the limits of its use,” he said.
International scholars interviewed by AJP offered differing views on where those limits should lie — in the technology itself, the arguments it produces or the institutions that hold leaders accountable.
Lawrence Lessig, the Roy L. Furman Professor of Law and Leadership at Harvard Law School, challenged the notion that citing AI in a dispute over executive authority was inherently problematic.
“I do not think it is democratically problematic to cite AI’s interpretation in a dispute over executive authority. It is only problematic if the argument cited is a bad one,” Lessig said.
The law and technology scholar compared AI assistance with the professional advice politicians have long received from lawyers and other experts.
“Public officials do not need to declare which lawyers advise them in doing their work before AI. They do not need to declare what the political views of those lawyers are or what law review articles those lawyers read,” Lessig said.
Asked what constitutional rule should govern presidential use of generative AI, he rejected the need for technology-specific restrictions.
“I don’t think there should be a constitutional democracy rule for the use of technology in doing the work of a government official,” he said.
“I think the rule should be that the government official should do their work well. We should test their work. We should evaluate whether it is true, whether its arguments are supported.”
Simon Chesterman, the David Marshall Professor of Law at the National University of Singapore, drew a sharper distinction between consulting a system and conferring authority on it.
“Used carefully, AI can be another source of advice, but it cannot become a source of authority: an elected leader remains accountable for the judgment exercised, especially where the system being consulted is privately designed, trained, and operated,” Chesterman noted.
His concern extends beyond whether a chatbot gets a particular answer right. It also involves the private companies that build the systems on which governments increasingly rely.
Chesterman examines that shift in his paper “Silicon Sovereigns: Artificial Intelligence, International Law, and the Tech-Industrial Complex.”
“My concern with what I have called ‘silicon sovereigns’ is precisely that political power is already shifting towards a small number of unelected companies that control the infrastructure, models, and information environment on which governments increasingly depend,” he said.
If officials outsource judgment as well as infrastructure to those companies, he warned, political power could move beyond the reach of the mechanisms designed to hold it accountable.
“The more powerful and important their decisions, the more important it is that they are immersed in a network that holds that power to account,” Bryson said.
Her research, “Human-Centred Control of AI Is Both Possible and Necessary,” examines how people can retain meaningful oversight of AI systems, including in sensitive security settings.
“There’s no simple way to surveil for good use. We need to be vigilant in ensuring our political systems are safe and effective.”
For a president invoking an AI oracle, that scrutiny must also reach beyond the chatbot’s answer to the evidence behind it.
“I believe it is always important to say where knowledge derives from, so people can check your sources and ensure they agree with your interpretation,” Bryson said.
Copyright ⓒ Aju Press All rights reserved.