South Korea is investigating whether AI played a role in recent cyberattacks that hit more than seven banks and affected over 68,000 people. President Lee Jae Myung warns not to underestimate how AI can lower the barrier for hackers, noting that an open-source Chinese tool called Artex AI is believed to have been used. The attacks stole sensitive personal data, including annual income and loan limits, and have prompted officials to urge financial institutions to strengthen their defenses against AI-powered threats.
Meanwhile, Zscaler CEO Jay Chaudhry warns not to trust AI agents too much, cautioning they can make unplanned mistakes, leak sensitive data, and invite cyberattacks. He points to a Deloitte survey showing about 80% of organizations lack mature governance for agentic AI. Chaudhry plans to release a product roadmap to address these risks, describing rogue agents as the biggest threat because they operate at machine speed without breaks.
In the sports technology space, NVIDIA launched new AI products at the IBC conference in Amsterdam, including 3D Body Pose for athlete tracking, Video Frame Generation for slow-motion replays, and TrueHDR for video enhancement. The company is partnering with Hawk-Eye and Stats Perform to speed up innovation in sports broadcasting, building on research originally developed for autonomous vehicles.
On the productivity front, a Reddit user claims he and his partner sold around 200 websites in 12 months using AI-driven automation. They relied on tools like Apollo for leads, Claude Code for development, and Soro for SEO blogging. The story highlights how web design is shifting from manual coding to managing automated systems, with a team of just two people handling large-scale work.
Netskope launched the AgentSkope Marketplace, a single console for security teams to browse, deploy, and manage AI agents. The platform uses a unified catalog and pool of credits with a usage monitoring dashboard. A 2026 survey found 42% of security leaders said alerts went uninvestigated due to scattered signals, making this marketplace a practical response to growing complexity in security operations.
Researchers at NYU Langone developed an AI tool that may detect heart transplant rejection by analyzing electrocardiograms and blood tests. The tool identified rejection in 92% of cases compared to 71% for traditional biopsy methods, based on a study of 200 patients at risk and a control group of 100 patients. The AI uses machine learning to find patterns that signal rejection and alerts doctors to act sooner.
CFOs are moving financial approval outside AI agents to prevent agents from altering transactions after approval. The recommended approach is to ensure no single system handles requesting, approving, and executing money movement. AI agents connected to email, procurement software, ERP systems, and bank accounts can accidentally bypass segregation of duties controls, so risk-based autonomy with stronger controls for unusual transactions is advised.
Penumbra, a new tool created by researchers, tests AI agents in finance, healthcare, and law by using an adversarial search method to find pairs of answers where one follows the rules and the other breaks them. Testing showed Penumbra found 144 such pairs for financial rules and 49 pairs for medical triage rules, helping ensure AI systems are safe before deployment in critical industries where rule violations carry real penalties.
Key Takeaways
- <ul><li>South Korea investigates AI-powered bank hacks affecting 68,000 people across 7+ financial institutions, with President Lee Jae warning about lowered hacking barriers</li><li>Zscaler CEO warns not to trust AI agents, citing that 80% of organizations lack mature governance for agentic AI</li><li>NVIDIA launches sports AI products including 3D Body Pose, Video Frame Generation, and TrueHDR at IBC Amsterdam</li><li>A web designer sold 200 websites in one year using Claude Code, Apollo, and Soro with a team of just two people</li><li>Netskope launches AgentSkope Marketplace for security teams to manage AI agents, addressing 42% of leaders reporting uninvestigated alerts</li><li>NYU Langone AI tool detects heart transplant rejection in 92% of cases vs. 71% for traditional biopsy methods</li><li>CFOs are moving financial approval outside AI agents to prevent unauthorized transaction alterations</li><li>Penumbra tool finds 144 rule-breaking pairs in financial AI and 49 in medical triage AI, helping validate safety before deployment</li><li>Artex AI, a Chinese open-source security tool, is believed to have been used in the South Korea bank attacks</li><li>Safeline offers six months of free AI wellness support to training clients and return-to-work participants</li></ul>
South Korea bank hacks show how AI could worsen cyberattacks
South Korea is investigating whether AI was used in recent cyberattacks on more than seven banks. Over 68,000 people were affected by the breaches. President Lee Jae Myung said AI could make hacking easier for people without special skills. A tool called Artex AI, an open-source Chinese security testing tool, is believed to have been used. Police have launched an investigation but have not confirmed AI carried out the attacks.
South Korea bank hacks warn about AI-powered cyberattacks
Some of South Korea's biggest banks were hacked using an open-source AI tool called Artex AI. The tool was developed in China and was originally designed to find cybersecurity vulnerabilities. At least seven financial firms were targeted, and personal information for 68,000 people was stolen. This includes data like annual income and loan limits. Officials are urging banks to improve their defenses against AI-powered threats.
Cybersecurity CEO warns not to trust AI agents
Zscaler CEO Jay Chaudhry warns that companies and individuals trust AI agents too much. He says AI agents can make unplanned mistakes, leak sensitive data, and invite cyberattacks. About 80% of organizations lack mature governance for agentic AI according to a Deloitte survey. Chaudhry plans to share a product roadmap to address AI agent risks. He describes rogue agents as the biggest risk because they work at machine speed without breaks.
NVIDIA expands sports technology with AI products
NVIDIA launched new AI media and pose estimation products at the IBC conference in Amsterdam. The tools include 3D Body Pose for athlete tracking, Video Frame Generation for slow-motion replays, and TrueHDR for video enhancement. NVIDIA is partnering with companies like Hawk-Eye and Stats Perform to speed up innovation in sports broadcasting. The technology builds on research originally developed for autonomous vehicles. The goal is to help developers and leagues improve digital experiences in sports.
Web designer sells 200 websites using AI in one year
A Reddit user claims he and his partner sold around 200 websites in 12 months using AI-driven automation. They used tools like Apollo for leads, Claude Code for development, and Soro for SEO blogging. The user says AI helped automate finding client problems, building websites, and generating content. The story shows how web design work is shifting from manual coding to managing automated systems. He credits AI for allowing a team of just two people to handle large-scale work.
Netskope launches AI agent marketplace for security teams
Netskope announced the AgentSkope Marketplace, a single console for security teams to browse, deploy, and manage AI agents. The marketplace uses a unified catalog and pool of credits with a usage monitoring dashboard. It is available now to all Netskope customers. A 2026 survey found 42% of security leaders said alerts went uninvestigated due to scattered signals. The marketplace adds to Netskope One AgentSkope, which launched in early 2026.
AI May Help Catch Heart Transplant Rejection Without Biopsies
NYU Langone researchers developed an AI tool that may detect heart transplant rejection by analyzing electrocardiograms and blood tests. The tool identified rejection in 92% of cases compared to 71% for traditional biopsy methods. The study included 200 patients at risk of rejection and a control group of 100 patients. The AI uses machine learning to find patterns that signal rejection and alerts doctors to act.
How CFOs Are Building Approval Controls Finance Agents Cannot Break
CFOs are moving financial approval outside AI agents to prevent agents from altering transactions after approval. The key rule is that no one should be able to request, approve, and execute money movement in a single system. AI agents connected to email, procurement software, ERP systems, and bank accounts can accidentally bypass segregation of duties controls. The recommended approach uses risk-based autonomy with stronger controls for unusual transactions.
Six months of free AI wellness support for Safeline staff
Safeline can offer its platform to training clients and return-to-work participants for six months of free AI wellness support. Use of the platform is voluntary. Engagement data does not go to WSIB or inform claims.
AI ethicist visits Austin Public Schools
An AI ethicist visited Austin Public Schools to teach students and staff how AI can be used in education. The event was designed to improve understanding of AI tools in school settings. The visit took place on October 5, 2026.
Penumbra finds hidden rule breaks in AI answers
Researchers created a new tool called Penumbra to test AI agents in finance, healthcare, and law. These sectors face real penalties if an AI violates rules, even when the mistake is hard to spot. Penumbra uses an adversarial search method to find pairs of answers where one follows the rules and the other breaks them. The tool only changes a few words to show exactly where the AI fails to meet an obligation. Testing showed Penumbra found 144 such pairs for financial rules and 49 pairs for medical triage rules. This approach helps ensure AI systems are safe before they are used in critical industries.
Sources
- South Korea bank hacks: The hidden third-party risk AI could make worse | Decoded
- South Korea’s bank hacks offer a warning about AI-powered cyberattacks
- Don't trust AI agents, warns the CEO of cybersecurity powerhouse
- NVIDIA Expands Sports Technology Ecosystem with Advanced AI Media and Pose Estimation Products
- I sold 200 websites in 12 months: Web designer claims how AI helped team of 2
- Netskope launches AI agent marketplace for security teams
- AI May Help Catch Heart Transplant Rejection Without Biopsies
- How CFOs Are Building Approval Controls Finance Agents Can’t Break
- Six months of free AI wellness support for Safeline staff
- A.I. ethicist visits Austin Public Schools
- Penumbra: Sample-Efficient Adversarial Search for Regulatory Obligations
Comments
Please log in to post a comment.