South Korea’s tax accountant exam scandal has raised questions about how to prevent AI-assisted cheating. Han Won-seok, 34, admitted that he used a miniature camera to photograph questions, sent them to an AI tool and checked answers on his smartphone. He ranked first in both rounds of the 63rd exam, scoring 397.5 out of 400 in the first round, and also admitted using the method in the first stage of the certified public accountant exam. Seoul police are investigating. One report separately says the Korea Institute of Public Finance caught an unnamed test-taker answering 90% of questions with a tax-focused AI tool; the reports do not establish whether that incident involved Han. Officials have also cited a separate case in which an AI inspector caught smart glasses during a gas engineer exam.
Companies are drawing boundaries around outside AI agents. United Airlines will block unauthorized agents from using its digital platforms, citing customer-account security, personal information and safety checks that passengers must complete themselves. JPMorgan Chase, eBay and Amazon are also restricting access, with eBay and Amazon limiting Meta’s Muse assistant. JPMorgan may permit approved API access, while United says booking services need permission. The approaches vary: Delta and American allow some AI-assisted bookings with safeguards, and Southwest offers a ChatGPT flight-shopping plugin. United says it could reconsider as assistants and protections develop.
Meta has taken seven months to develop Muse, its personal AI agent, because, product lead Mona Sarantakos said, a failure could stop users from returning. Muse can work across a person’s inbox, calendar and the web from a short prompt. Nat Friedman developed Sentinel as a safety backstop, and the team expanded testing before release. Sarantakos says possible future hardware includes glasses and charm devices. Separately, a Crissy Field review of 34,545 sites using Supabase found that 24,664 returned database records on first load; email addresses appeared in tables across 7,403 sites. Findings included sensitive records, but the review cautioned that this does not confirm leaks at 30% of sites, since some data may be public or test material. Missing Row Level Security rules were a recurring issue.
AI investment prospects remain difficult to judge, according to Oaktree co-founder Howard Marks. He said AI could require an average of $1 trillion in new capital annually, while AI-related debt issuance had neared $600 billion globally this year, citing Goldman Sachs data. Marks did not call AI a bubble, but questioned whether current leaders will stay ahead and whether models could become commodities. Another article presents Anthropic’s planned November 19 IPO, described as roughly six times oversubscribed at a $1.4 trillion valuation, as a potential lift for the AI trade. It also flags possible short-term financing and supply-chain risks associated with Jensen Huang’s departure, while arguing that big tech balance sheets and cloud growth support longer-term investment. Nvidia competitors could gain share.
Electricity and construction timelines are a further constraint on data-center expansion. Former Nvidia CEO Jensen Huang estimated that a 1-gigawatt data center costs about $50 billion to build and could produce around $40 billion in annual revenue; a three-month delay could mean $10 billion in lost revenue potential. The article highlights power supply concerns and points to NuScale Power and Oklo as higher-risk small modular reactor plays, and Constellation Energy and Bloom Energy as more risk-conscious options. Constellation has signed power purchase agreements with major AI operators, while Bloom makes fuel-cell systems for on-site power.
Other articles focus on people and strategy behind AI’s growth. Startups and venture firms are hiring ecosystem roles to organize events, partnerships, customer support and training; Clay’s Natalie Sportelli, for example, has led product demonstrations, customer coaching and an AI-sales panel. GeekWire’s Etzioni argues that faster feature development is not a substitute for strategy: founders should test cheaply, scale only after experiments work, and build defensibility such as proprietary data access through trusted relationships. In Hollywood, SAG-AFTRA President Sean Astin is urging Congress to pass the No Fakes Act, which would let people control digital replicas of their voices and likenesses and establish accountability. The bill has not come to a vote; Astin also supports a federal production tax credit and opposes the Paramount-Warner Bros. merger, citing workers’ interests.
Key Takeaways
- Han Won-seok admitted using a miniature camera, AI and a smartphone to cheat on South Korea’s tax accountant exam; he scored 397.5/400 in the first round and ranked first in both rounds. Police are investigating.
- Han also admitted using the same method in the first stage of the certified public accountant exam. A separate report describes an unnamed tax-exam taker answering 90% of questions with AI.
- United will restrict unauthorized AI agents from booking through its platforms, citing account security, privacy and passenger safety checks. JPMorgan Chase, eBay and Amazon are also limiting outside agents; eBay and Amazon have restricted Meta’s Muse.
- Meta spent seven months developing Muse, which can work across a user’s inbox, calendar and the web. Nat Friedman developed Sentinel as a safety backstop.
- A Crissy Field review found that 24,664 of 34,545 sites using Supabase returned database records on first load; exposed material included sensitive data, but the review said this does not confirm leaks at 30% of sites.
- Howard Marks said AI may require about $1 trillion in new capital annually and cited nearly $600 billion in AI-related global debt issuance this year, while warning that future winners and model economics are hard to predict.
- One article describes Anthropic’s planned November 19 IPO as roughly six times oversubscribed at a $1.4 trillion valuation; it also flags potential near-term financing and supply-chain risks and possible gains for Nvidia competitors.
- Jensen Huang estimated a 1-gigawatt data center costs about $50 billion and could generate $40 billion annually; a three-month delay could mean $10 billion in lost revenue potential. Power options discussed include Constellation Energy and Bloom Energy.
- AI startups and venture firms are adding ecosystem jobs focused on relationships, partnerships, events and customer support. An analysis argues startups need strategy, careful scaling and defensibility—not just faster development.
- SAG-AFTRA President Sean Astin is seeking passage of the No Fakes Act to protect control of digital voice and likeness replicas; the bill has not yet come to a vote.
South Korean tax exam’s top scorer admits using AI to cheat
The top scorer in South Korea’s national tax accountant exam admitted he used AI to get answers during the tests. He ranked first in both rounds, held in April and July, and scored 397.5 out of 400 in the first round, missing only one of 160 questions. He first denied accusations that he used smart glasses, but later submitted a written confession through his lawyer. Police opened an investigation.
Han Won-seok admits using a camera and AI to cheat on exams
Han Won-seok, the top scorer in South Korea’s tax accountant qualification exam, admitted that he used a miniature camera, AI and a smartphone to cheat. He photographed questions and checked AI’s answers on his phone, earning 397.5 out of 400 in the first round and 663 out of 800 in the second. Han also admitted using the same method in the first stage of the certified public accountant exam. Seoul police opened an investigation, and Han said he would cooperate. He said financial struggles and anxiety about passing drove him to cheat.
South Korean exam top scorer confesses to AI-assisted cheating
Han Won-seok, 34, admitted that he used a micro-camera to photograph questions during South Korea’s state-run tax accountant exam, then sent them to AI and checked the answers on his smartphone. He said he had turned himself in to the Seoul Metropolitan Police Agency, which opened an investigation. His near-perfect score drew online scrutiny after the exam results were announced on Oct. 28. The report says about 10 percent of applicants passed the exam and describes Han’s confession as the first known case of a successful candidate admitting to using such equipment to cheat.
Report says unnamed tax exam taker used AI to answer questions
The report says the Korea Institute of Public Finance caught an unnamed person using AI to answer questions on South Korea’s tax accountant exam. It states the person answered 90 percent of the questions correctly with an AI tool designed for tax-related questions. The incident has raised concerns about AI use in exams and the need for stronger steps to prevent cheating. The institute said it would take action and provide more information in the coming days.
South Korean police investigate top tax exam scorer after confession
South Korean police opened an investigation after the top scorer in the national tax accountant exam admitted to using AI to cheat. The candidate ranked first in both rounds, held in April and July, and scored 397.5 out of 400 in the first round. He said he used a miniature camera to photograph questions, sent them to AI and checked answers on his smartphone. The Human Resources Development Service of Korea administers the exam, and the report also says an AI inspector caught someone wearing smart glasses during a gas engineer certification exam last month. South Korea’s Suneung university entrance exam was scheduled for Nov. 19, 2026, as officials discussed ways to address cheating with smart devices.
South Korea tax exam top scorer admits using AI to cheat
A candidate who scored 397.5 out of 400 on South Korea’s tax accountant exam admitted to cheating, just days before the national college entrance exam. He said he used a miniature camera to photograph questions, sent them to a generative AI tool and checked the answers on his smartphone. The candidate, identified as Han, ranked first in both exam rounds held in April and July. Seoul police received his written confession and assigned the case to investigators. The scandal has prompted government officials to prepare measures to protect the fairness of the college entrance exam.
South Korea tax exam top scorer confesses to AI-assisted cheating
The top scorer in South Korea’s 63rd certified tax accountant qualification exam admitted on November 15, 2026, that he cheated in both exam stages. He said he used a miniature camera to photograph questions, then sent them to generative AI and checked the answers on a smartphone. His score included 159 correct answers out of 160 in the first stage. The Seoul Metropolitan Police Agency received his written confession and assigned the case to investigators. Reports do not confirm which AI tool he used or the camera model.
Police investigate South Korea exam leader after AI cheating confession
South Korean police began investigating the country’s top tax accountant exam scorer on November 15 after he admitted using AI to cheat. He said he photographed questions with a miniature camera, sent them to AI and checked the answers on his smartphone. He scored 397.5 out of 400 in the first round, with 159 correct answers out of 160, and ranked first in both rounds held in April and July. He also said he had struggled financially after leaving his job to start a business. The Human Resources Development Service of Korea said an AI inspector had caught someone wearing smart glasses during a gas engineer exam the previous month.
South Korean tax exam leader admits cheating with AI
Han Won-seok, 34, admitted in a video posted on attorney Lee Don-ho’s YouTube channel that he used a miniature camera to photograph tax accountant exam questions, sent them to an AI tool and checked answers on a smartphone. He said he had submitted a written confession to Seoul police and also admitted using the same method to cheat in the first round of the certified public accountant exam in March. Han scored 397.5 out of 400 in the first round of the 63rd tax accountant exam and ranked first in its written second round. This year, 722 of the 6,606 people who took the second round passed, a pass rate of 10.92 percent. Authorities and professional groups had called for an investigation after his unusually high scores raised suspicions.
United blocks outside AI agents from booking flights
United Airlines will restrict unauthorized third-party AI agents from using its digital platforms to book flights, including through travel agencies and booking sites. The airline says the rules protect customer accounts and personal information, and warn that AI tools could make unwanted purchases or change account details. United also says passengers must personally confirm some safety information during check-in, which an AI agent cannot do for them. The policy makes United the first major US airline to clearly announce such restrictions, while Delta and American allow some AI-assisted bookings with safeguards, and Southwest has a ChatGPT flight-shopping plugin. United says it may reconsider its position as AI assistants and protections develop.
Major Companies Block Outside AI Agents Over Security and Customer Control
JPMorgan Chase, United Airlines, eBay and Amazon are limiting outside AI agents from accessing their systems. JPMorgan CEO Jamie Dimon said unauthorized agents pose risks, while United cited account security and aviation safety. The companies also worry that AI assistants could take over customer relationships. JPMorgan may allow access through approved APIs, and United says AI booking services must get permission first. eBay and Amazon have also restricted Meta’s AI assistant Muse.
Research Finds Lovable Apps Exposed Passports, Messages and Other Data
A Crissy Field review found that many websites built with Lovable and Supabase exposed database records without requiring a login. Among 34,545 sites using Supabase, 24,664 returned records on first load, and the review found email addresses in tables across 7,403 sites. Exposed information included profiles, chat messages, ticket records, passport details, contracts and invoices. Crissy Field said the findings do not mean that 30% of the sites had confirmed leaks, since some records may be public or test data. The report said missing Row Level Security rules were a recurring cause of exposure.
Howard Marks Says AI Investment Returns Remain Hard to Predict
Howard Marks, co-founder of Oaktree Capital Management, warned that investors cannot easily tell which companies will profit from artificial intelligence. He said AI could require an average of $1 trillion in new capital each year, while AI related debt issuance had neared $600 billion globally this year, according to Goldman Sachs data cited in his memo. Marks questioned whether today’s AI leaders will remain ahead and whether models may become commodities. He stopped short of calling AI a bubble, but urged investors to assess risks carefully and recognize how little they can predict.
Anthropic IPO Could Help Renew Confidence in the AI Trade
The article says Jensen Huang’s departure could create short term risks for AI financing and supply chain coordination, possibly slowing data center spending. It presents Anthropic’s planned November 19 IPO, described as roughly six times oversubscribed at a $1.4 trillion valuation, as a possible boost for the AI trade. The article says the longer term AI buildout remains supported by large technology companies’ balance sheets and cloud growth. It also says Nvidia competitors could gain market share. The author says three scenarios will shape the outcome and describes keeping investment decisions flexible and maintaining liquidity.
Power Delays Could Cost Data Centers Billions in Lost Revenue
Former Nvidia CEO Jensen Huang said a 1 gigawatt data center costs about $50 billion to build and could generate around $40 billion in yearly revenue. He estimated that a three month construction delay could mean $10 billion in lost revenue potential. The article identifies power supply as a major limit on building more AI data centers, citing concerns from McKinsey, Morgan Stanley, J.P. Morgan and Goldman Sachs. It discusses small modular reactor companies NuScale Power and Oklo as higher risk options, and names Constellation Energy and Bloom Energy as more risk conscious choices. Constellation has signed power purchase agreements with major AI operators, while Bloom makes fuel cell systems that can help data centers generate electricity.
AI startups are hiring people to build human connections
Startups and venture capital firms are adding jobs focused on building relationships across their networks. These workers plan events, set up partnerships, support clients and lead training sessions. Yoni Rechtman of Slow Ventures said an ecosystem includes customers, partners, distribution channels, supporters and advocates. Natalie Sportelli, Clay’s start-up ecosystem lead since January, has organized product demonstrations, customer coaching and a London panel on AI-focused sales teams.
Meta delayed Muse to build trust in its AI agent
Mona Sarantakos, Muse’s head of product, said Meta spent seven months building the personal agent because users may not return after it fails once. Muse can use a short prompt to work across a person’s inbox, calendar and the web, such as creating a school week brief. Sarantakos said the team began work after Nat Friedman contacted a small group about OpenClaw near the end of January, and Mark Zuckerberg gave feedback as the team tested the product. Nat Friedman developed Sentinel as a safety backstop, while the team expanded testing until it had avoided a catastrophic failure. Sarantakos said future developments to watch include whether Muse moves beyond its app to glasses and charm devices.
Sean Astin pushes for AI protections for Hollywood performers
SAG-AFTRA President Sean Astin is working to protect the union’s about 160,000 members as artificial intelligence and shrinking work reshape Hollywood. He is urging Congress to pass the No Fakes Act, which would let people control digital copies of their voices and likenesses and create a path for accountability. The bill has not yet come to a vote, and Astin has made frequent trips to Washington to build support. He has also backed a federal tax credit for U.S. film and television productions and spoken against the Paramount-Warner Bros. merger, which he says is bad for workers. Astin wants to slow AI’s use in the industry and says changes are needed at the state and national levels.
AI startup speed cannot replace a clear strategy
AI tools can help startups build features faster, but GeekWire’s Etzioni argues that speed alone is not a strategy. He says competing products can appear quickly, new model releases can reduce the value of some data, and growth driven by hype can limit a company’s choices. Startups should run cheap experiments quickly, then scale more carefully because hiring, contracts and compute purchases are harder to reverse. Sri Chandrasekar of AI House says founders need judgment to find the right problem and should build a full system only after an experiment works. The article argues that durable businesses need defensibility, such as access to proprietary data through trusted relationships.
Sources
- Top tax accountant exam scorer in South Korea admits to using AI to cheat
- Top tax accountant exam scorer admits to cheating using AI
- Korea sees first AI cheating admission by successful state exam candidate
- Breaking: Top Tax Accountant Exam Scorer Admits Cheating with AI
- South Korea probes top exam scorer after AI cheating confession
- AI cheating scandal rattles South Korea as annual college exam looms
- Tax Exam AI Cheating: Top Scorer Admits 159/160 Score
- South Korea probes top exam scorer after AI cheating confession
- Tax accountant exam's top scorer admits to cheating with AI smart glasses
- United Airlines declares war on AI travel agents - in a major blow to Meta's plans to let chatbots book your holidays
- JPMorgan, United Airlines Block Outside AI Agents: Security Cited, Control at Stake
- Lovable apps left passports and chats exposed
- Howard Marks Warns AI Investing Is a $1 Trillion Gamble: 'Practically Unknowable' Profits
- Only Anthropic Can Save The AI Trade Now
- Nvidia's Ousted CEO Said a 1-Gigawatt Data Center Delay Costs $10 Billion a Quarter. Here's Why That Matters for Power Stocks.
- In a Sloppified A.I. World, Only Humans Can Build ‘Ecosystems’
- Meta Held Muse to Make Trust Stick
- SAG-AFTRA President Sean Astin's toughest role yet: trying to tame AI
- Etzioni on AI: Startup speed is not a strategy
Comments
Please log in to post a comment.