The head of South Korea's guard consulted ChatGPT before martial law was imposed

Context: South Korean Martial Law Crisis

• Commenters clarify that this refers to the 2024 South Korean martial law crisis, which lasted only about six hours but triggered a major political fallout.
• The affair is now treated as an attempted coup/treason, with an ongoing impeachment process against the president and investigations into who knew what, and when.
• One view stresses that South Korea remains a robust democracy precisely because attempts to seize power are punished; another notes that many former presidents have faced legal trouble, which might motivate desperate behavior by incumbents.

Role of ChatGPT and Prior Knowledge

• The key point from the article (via translation) is that the head of the presidential guard queried ChatGPT about what to do in case of martial law before cabinet members had arrived and hours before the declaration.
• This contradicts his reported claim that he only learned of martial law from TV, suggesting prior involvement or knowledge.
• Several commenters emphasize that the scandal is about his apparent lie and role in the plot, not about AI “advising” on martial law.

AI Influence on Decisions and Trustworthiness

• Some argue that leaders almost certainly already use LLMs for day‑to‑day advice, and that this might even be better than relying on biased or ill‑intentioned humans.
• Others strongly counter that LLMs are highly fallible, hallucinate confidently, lack transparency and diversity of viewpoints, and are more dangerous than sources like Wikipedia.
• A recurring concern is that people over-trust computer outputs and that LLMs rarely admit ignorance, short‑circuiting users’ skepticism.

AI in Warfare

• Commenters cite military AI platforms and AI‑assisted targeting (e.g., in Gaza) as evidence that “AI in war” is already here and disturbing.
• There is sharp disagreement over claims of precision, civilian casualties, and the ethics of such systems, with no consensus.

Privacy, Surveillance, and Regulation

• Many criticize officials for using a US-hosted LLM to research a sensitive, non‑public operation, noting that anything typed into an online textbox is potentially visible to third parties and discoverable in forensics.
• Discussion turns to GDPR, cookie banners, and “performative” privacy regulation—some see GDPR as effective and culture‑shifting; others see weak enforcement and dark‑pattern compliance.
• Several propose stronger consent and data-sovereignty frameworks where individuals can see and control who accesses their data.