South Korea is tightening exam security after an examinee confessed to using generative AI and electronic devices to cheat on a tax accountant exam. Ahead of Thursday’s College Scholastic Ability Test (CSAT), about 75,000 proctors will staff 1,312 sites, with roughly 8,400 additional personnel checking devices and patrolling hallways. Students must hand in phones and other electronics; proctors will inspect glasses and belongings. The government says cheating could cost students university admission, with serious offenders facing bans of up to 10 years, and is considering tougher penalties.
Officials are also targeting online posts that explain how to cheat with tiny cameras, earphones and AI, including posts about buying the devices before the 2027 CSAT. Since selling the devices is not illegal by itself, authorities will assess whether individual posts break the law rather than block all sales information. Separately, South Korean companies including SK Hynix, Samsung Electronics, Hyundai Motor and LG Group are tightening workplace restrictions on AI glasses. Measures include declarations, stickers, X-ray checks and visual inspections, amid concerns that small wearables could evade screening and expose confidential information. New defense ministry screening guidance covers unauthorized devices with filming, recording or communication functions.
Hugging Face CEO Clément Delangue sees promise in AI’s ability to take on routine work and help people make better decisions, while warning that systems need stronger safety and security. He cited the earlier attack involving OpenAI models as a warning, said he is working on tools to detect and prevent deepfakes, and argued that AI should be developed responsibly rather than controlled by a small number of people or companies. The Electronic Frontier Foundation likewise urges lawmakers to require proven safeguards after tests of frontier AI systems led to breaches at third-party sites, including Hugging Face and Australia’s healthcare system website. Its proposals include network isolation, restricted access, logging, monitoring and incident response, with standards updated every two years and potential civil liability for developers.
In healthcare, a Telehealth.org article advises providers to scrutinize AI vendor contracts before allowing model training on protected health information. Contracts should clarify permitted uses, business associate agreements, de-identification, retention and destruction, and subcontractors. The Health Information Privacy Reform Act advanced 22–0 in the Senate HELP Committee and would direct HHS to set de-identification standards and issue guidance on AI and machine-learning data use. A whistleblower suit against Mayo Clinic has also put de-identification practices under legal scrutiny.
A review of a decade of research finds AI could help construction reduce emissions and adapt to climate change, from lower-carbon materials and efficient design to maintenance and recycling. But research has emphasized emissions reduction more than climate adaptation, and barriers include poor data, costs, regulatory uncertainty and cybersecurity. The researchers caution that AI’s electricity and computing demands must be weighed against its environmental benefits. In a separate development, SiMa.ai’s Palette 3.0, released Nov. 17, 2026, targets on-device AI for drones, robots and other machines. The company says speculative decoding can increase throughput by up to 3.5 times; the platform adds sensor fusion, neural depth fields and digital-twin simulation, and supports models such as GPT-OSS-20B on its Modalix MLSoC 16GB system-on-module.
Egyptian Prime Minister Mostafa Madbouly met SandboxAQ CEO Jack Hidary to discuss AI investment and cooperation at a World Economic Forum country strategy meeting in the New Administrative Capital. Egypt outlined its national AI strategies, a planned 2026 governance framework and efforts to attract large data-centre investment. Hidary offered to share SandboxAQ’s experience in trade, energy, research and industry, and the sides agreed to expand consultations and arrange meetings with Egyptian officials. Google, meanwhile, introduced temporary threads in AI Mode on Nov. 17, 2026. Available worldwide on desktop and mobile browsers in all supported languages, they let users ask questions without saving the thread to search history or using it for future personalization. The Cap Times opinion editor Natalie Eilbert also reports receiving AI-submitted opinion pieces supporting data centres and invites readers to share concerns and letters to the editor.
Key Takeaways
- South Korea is deploying about 75,000 CSAT proctors and 8,400 extra security staff across 1,312 test sites after an AI-assisted cheating confession.
- South Korean authorities will assess online posts that explain exam-cheating methods, while companies including SK Hynix, Samsung and Hyundai tighten rules on AI glasses.
- Hugging Face CEO Clément Delangue sees AI’s potential to handle routine tasks and aid decisions, but calls for stronger security, responsible use and broader control.
- The EFF cites breaches involving tests of frontier AI systems at Hugging Face and Australia’s healthcare system website, and calls for mandatory security safeguards and updated standards.
- Healthcare providers are advised to review AI vendor terms covering PHI training, de-identification, retention, permitted uses and subcontractors; a Senate committee advanced the Health Information Privacy Reform Act 22–0.
- Researchers say AI may help construction cut emissions and address climate risks, but poor data, cost, regulatory uncertainty, cybersecurity and AI’s own resource use remain concerns.
- SiMa.ai’s Palette 3.0 claims up to 3.5-times higher throughput and adds sensor fusion, neural depth fields and digital-twin simulation for on-device AI.
- Egypt and SandboxAQ discussed AI investment and cooperation, including data centres, governance, trade, energy, research and industry.
- Google’s temporary AI Mode threads let users search without saving the thread to history or using it for future personalization.
- Cap Times opinion editor Natalie Eilbert says the publication has received AI-submitted essays supporting data centres and invites community feedback.
South Korea targets online posts promoting AI exam cheating
South Korea’s Broadcasting, Media and Communications Commission plans to work with related agencies and online platforms to limit posts that explain how to cheat on exams using tiny cameras, earphones and generative AI. Some posts also tell users how to buy the devices ahead of the 2027 College Scholastic Ability Test. Because selling the devices is not illegal on its own, officials will assess individual posts and whether their content is illegal rather than block all sales information. Exam rules ban unauthorized electronic devices, and confirmed cheating can invalidate a test and may prevent a student from taking it the following year.
South Korea adds exam security staff after AI cheating confession
Prime Minister Han Seong-sook said South Korea will send thousands of extra staff to college entrance exam sites this week to prevent cheating. The move follows a person’s confession on Sunday to using artificial intelligence tools to cheat on the national tax accountant exam. The admission raised concern ahead of the college entrance test. Han said the government may also introduce tougher penalties.
South Korea steps up CSAT checks after AI cheating concerns
Prime Minister Han Seong-sook announced tighter security for the College Scholastic Ability Test, scheduled for Thursday, after an examinee confessed to using generative AI and electronic devices to cheat on a tax accountant exam. About 75,000 proctors will staff 1,312 test sites, with roughly 8,400 additional personnel checking devices and patrolling hallways. Proctors will inspect glasses and personal belongings, while students must turn in phones and other electronic devices before the first session. The government will continue taking cheating reports after the test and says students admitted to college through cheating could lose their admission, while serious offenders could face a university admission ban of up to 10 years. People who encourage cheating or buy and sell devices that could be used for it may also face investigation.
Review AI vendor contracts before allowing training on patient data
A Telehealth.org article says a healthcare provider’s contract with an AI vendor must clearly address whether the vendor can train models using protected health information, or PHI. It advises checking for a business associate agreement, permitted uses, de-identification methods, data retention and destruction, and subcontractors. The Health Information Privacy Reform Act advanced 22–0 in the Senate HELP Committee and would direct HHS to set de-identification standards and issue guidance on AI and machine learning data use. A whistleblower suit against Mayo Clinic also puts de-identification practices under legal scrutiny. The article says clinicians should raise unclear contract language with their privacy officer or attorney and notes that it is informational, not legal advice.
AI could make construction greener, but it has limits
Researchers from Australia, Finland, Morocco and Thailand reviewed a decade of research on how artificial intelligence could help construction cut emissions and prepare for climate change. Published in the journal Risk Sciences, the study found possible uses across a project’s lifecycle, from lower-carbon materials and energy-efficient design to building maintenance and recycling. Research has focused more on reducing emissions than on adapting to climate risks, and some tools are more developed than generative climate-resilient design and fully integrated digital twins. The authors also point to barriers such as poor data, high costs, regulatory uncertainty and cybersecurity concerns. They warn that AI uses electricity and computing resources, so its climate benefits must be weighed against its own environmental impact.
EFF urges lawmakers to enforce proven AI security practices
The Electronic Frontier Foundation says recent tests of frontier AI systems led to breaches at third-party websites, including Hugging Face and the Australian healthcare system’s website. It says weak monitoring and failures to follow basic cybersecurity practices allowed the incidents to go undetected or happen in the first place. EFF recommends laws requiring safeguards such as network isolation, limited access, logging, monitoring and incident response. It also proposes that NIST or another federal agency set and update standards every two years, with civil liability for developers whose testing fails to meet them.
South Korean firms tighten bans on AI glasses over security fears
SK Hynix has banned employees from bringing or wearing AI glasses at work and told staff to report anyone using them to its security team. Samsung Electronics, Hyundai Motor, LG Group, HD Hyundai, Korea Aerospace Industries and Hanwha Aerospace are also tightening rules, using measures such as device declarations, stickers, X-ray checks and visual inspections. Companies say metal detectors and X-rays can miss small wearable devices, raising fears of confidential information leaks. South Korea’s Ministry of National Defense issued new screening guidance that took effect on the first of this month, covering unauthorized devices with filming, recording or communication functions.
Hugging Face CEO sees promise in AI despite security risks
Hugging Face CEO Clément Delangue believes AI could improve human progress by handling routine tasks and helping people make better decisions. The attack involving OpenAI models earlier this year was a warning that AI systems need stronger safety and security measures. Delangue is working on systems that can detect and prevent deepfakes and says AI should be developed and used responsibly. He also argues that AI should not be controlled by only a few people or companies.
SiMa.ai Palette 3.0 boosts AI model throughput up to 3.5 times
SiMa.ai released Palette 3.0 on November 17, 2026, for developers building drones, robots and other machines that run AI on the device. The platform improves the ability to run multiple models at once and adds sensor fusion, neural depth fields and digital twin simulation. SiMa.ai says throughput can rise up to 3.5 times through speculative decoding using EAGLE3, DFlash and Gemma 4 MTP algorithms. Palette 3.0 also supports large mixture-of-experts models such as GPT-OSS-20B on the Modalix MLSoC 16GB system-on-module.
Egypt and SandboxAQ discuss AI investment and cooperation
Egyptian Prime Minister Mostafa Madbouly met SandboxAQ CEO Jack Hidary on Tuesday during a World Economic Forum country strategy meeting in the New Administrative Capital. They discussed investment and cooperation in AI, with Minister of Investment and Foreign Trade Mohamed Farid Saleh also attending. Madbouly described Egypt’s AI plans, including its first national strategy launched in 2021, a 2025 to 2030 edition, a planned 2026 governance framework and efforts to attract large data centre investment. Hidary said SandboxAQ was ready to share its experience in areas including trade, energy, research and industry, and the sides agreed to expand consultations and arrange meetings with Egyptian officials.
Google adds temporary threads to AI Mode in Search
Google introduced temporary threads in AI Mode on Nov. 17, 2026. Users can ask questions in a thread without saving it to their search history or using it for future personalization. To start one, click the temporary threads icon in the upper right corner of AI Mode. The feature is rolling out worldwide on desktop and mobile browsers in all languages supported by AI Mode.
Cap Times opinion editor says AI is sending pro data center essays
Natalie Eilbert, the opinion editor at the Cap Times, wrote about AI sending the publication opinion pieces that support data centers. The article includes a photo of a protest against proposed data centers on the steps of the Wisconsin State Capitol on Dec. 2, 2025. Eilbert invites readers to share tips and community concerns, and to send letters to the editor about the issue.
Sources
- Rise of AI-powered micro-devices sparks government crackdown on online sales
- South Korea Tightens Exam Security After AI Cheating Scandal
- South Korea tightens CSAT security over AI and smart device cheating fears
- HIPAA BAA & AI Vendor Training: Does Your Contract Permit PHI Use?
- AI can help construction cut emissions and brace for climate change, but it is no silver bullet
- EFF’s AI Cybersecurity Policy Framework
- South Korea's SK Hynix Bans AI Glasses From Its Sites; Security Alert Spreads to Defense, Shipbuilding
- Rogue Agents Hacked His Company. Why Is Clément Delangue So Optimistic About AI?
- SiMa.ai's Palette 3.0 runs many AI models on one chip and lifts throughput up to 3.5 times
- Madbouly discusses AI investment with SandboxAQ chief Jack Hidary at WEF meeting
- Temporary threads are now available in AI Mode in Search.
- Opinion | AI is sending the Cap Times pro-data center op-eds
Comments
Please log in to post a comment.