Jensen Huang says artificial intelligence needs roughly a billion times more compute as reasoning and agent systems grow. He told attendees that inference costs will fall quickly, much like computing costs halved each year over past decades. Huang explained that harnesses give models working memory, knowledge graphs, and tool use so agents can connect to real company systems. Coding was the first area where agents gained traction, followed by design and creative work, though questions remain about safety verification and when revenue will catch up with capital spending.
Security researchers are showing how AI can both attack and defend systems. Cracken researcher Vladyslav Poddubnyi will present an autonomous red-team agent that uses live command and control telemetry to adjust attacks in real time. The agent moves into post-exploitation on its own and adapts to host configurations, privileges, and defenses without predefined scripts. Separately, European researchers demonstrated that AI can write personalized phishing emails for less than three cents each, and those emails were clicked nearly three times more than generic ones. In one case, Revolut handed passport images and transaction histories to someone pretending to be a government agency.
Other developments show AI spreading into telecom, education, and public health. Cerillion earned a spot in the 2026 Gartner Magic Quadrant for communications service provider AI tools, using model-agnostic assistants across billing, charging, and service fulfilment. In Connecticut, computer science enrollment has dropped by about 3,000 students since ChatGPT launched in 2022, even as lawmakers require AI topics in public schools. Teachers are using tools like Canva and Gemini to help English learners build independence while keeping thinking active. A University de Zaragoza-led consortium also received funding to model human behavior for tracking epidemics.
Federal and industry experts emphasize that AI works best when cybersecurity is built in from the start. The U.S. Army Corps of Engineers focuses on training workers and using digital twins to protect infrastructure like hydropower dams. Meanwhile, a panel on responsible AI stressed that human judgment should remain part of important decisions to protect agency and accountability.
Key Takeaways
- Jensen Huang says AI needs a billion times more compute as reasoning and agent harnesses expand
- Autonomous AI red-team agents can adapt attacks in real time using live command and control telemetry
- AI-generated phishing emails cost less than three cents each and are clicked nearly three times more than generic ones
- Revolut leaked passport images and transaction histories to a social engineer posing as a government agency
- Cerillion appears in the 2026 Gartner Magic Quadrant for CSP AI-enabled marketing and sales solutions
- Computer science enrollment in Connecticut fell by about 3,000 students since ChatGPT launched in 2022
- Teachers use Canva and Gemini to help English learners build independence while keeping active thinking
- University de Zaragoza leads a consortium modeling human behavior to track epidemic spread
- U.S. Army Corps of Engineers uses digital twins to protect infrastructure like hydropower dams
- Experts say human judgment must stay part of important AI-driven decisions
Jensen Huang says AI needs a billion times more compute
Jensen Huang stated that AI needs a billion times more compute as reasoning and agent harnesses grow. He noted that inference costs will fall quickly, comparing the drop to how computing costs halved yearly over past decades. He explained that the harness gives models working memory, knowledge graphs, and tool use so agents can connect to real company systems. Coding was the first area where agents gained traction, followed by design and creative work. The discussion left open how companies verify safety and when revenue will catch up with capital spending.
AI red-team agents use live C2 feedback to adapt attacks
Cracken researcher Vladyslav Poddubnyi will present research on an autonomous AI red-team agent that uses live command and control telemetry. The agent can generate and adjust its actions in real time based on what it observes. After initial execution, it moves into post-exploitation on its own. It adapts to the host's configuration, privileges, available tools, and defenses without relying on predefined scripts.
Exposed employee data fuels AI-powered social engineering
Researchers showed that AI can write personalized phishing emails for less than three cents each, and these emails were clicked nearly three times more than generic ones. European researchers collected work emails for 7,741 university employees and used a large language model to craft targeted messages. Revolut handed passport images and transaction histories to someone pretending to be a government agency. Trezor also suffered a phishing attack after a third-party email provider breach, showing how attackers use publicly available data and AI tools to build convincing pretexts.
Cerillion recognized in Gartner report for AI telecom tools
Cerillion was included in the 2026 Gartner Magic Quadrant for CSP AI-Enabled Marketing and Sales Solutions. The recognition covers AI tools that help communications service providers with marketing and sales. Cerillion uses AI assistants and agents with a model-agnostic architecture across product creation, service fulfilment, charging, and billing. Its AI Management Centre provides governance over AI model usage, and its Model Context Protocol Server enables controlled access to business systems. Chief executive Louis Hall said the recognition reflects investment in embedding AI into telecom software.
New AI project models human behavior to track epidemics
A consortium led by the University de Zaragoza has been selected for the 2026 Health Research Call. The project includes Universidad Carlos III de Madrid and focuses on modeling human behavior to track epidemic spread. The research aims to use AI to better understand how people move and interact during disease outbreaks. This approach could help public health officials respond more effectively to epidemics.
Preserving Human Judgment in the Age of AI
Experts say the main challenge of responsible AI is keeping human judgment in important decisions. Most panelists believe human judgment should stay part of human-machine systems. The goal is not to keep human judgment just for its own sake but to make sure it is available when it matters most. This helps protect human agency and accountability.
Computer Science Enrollment Drops in Connecticut After ChatGPT
Since ChatGPT launched in 2022, computer science enrollment in Connecticut has fallen each year. About 3,000 fewer students are taking these courses compared with the 2022-23 school year. Nationwide, four-year colleges saw a drop of 53,000 computer science majors in spring 2026. Connecticut lawmakers passed a new law requiring computer science, including AI topics, in public schools starting this year.
AI Success Requires Strong Cybersecurity Integration
Federal and industry experts say AI tools work best when cybersecurity is built in from the start. The U.S. Army Corps of Engineers focuses on accelerating AI use, training workers, and building trusted AI tools. It uses digital twins to protect infrastructure like hydropower dams. Experts also recommend starting with smart machine models and finding practical uses for AI in mission-critical work.
Using AI With English Learners While Keeping Thinking Active
Teachers are using AI tools like Canva and Gemini to help English learners build independence. Students create apps, set SMART goals, and practice spelling and vocabulary through repeated tweaks. Teachers also use AI to design culturally relevant examples and reduce bias in lessons. The aim is to encourage collaboration and keep students actively thinking instead of relying on AI to do the work for them.
Sources
- Huang: AI needs a billion times more compute
- Beyond Scripted Attacks: AI Red-Team Agents Driven by Live C2 Feedback
- How Exposed Employee Data Fuels AI-Powered Social Engineering
- Cerillion Recognised In Gartner Report For AI-enabled Telecoms Marketing And Sales
- New AI project models human behavior to track epidemic spread
- The Defining Challenge of Responsible AI Is Designing Systems That Preserve Human Judgment When and Where It Counts
- Is AI Leading to Declining Enrollment in Computer Science?
- Successful implementation of AI tools must start with integrating cyber capabilities
- How to Use AI With English Learners Without Encouraging Cognitive Offloading (Opinion)
Comments
Please log in to post a comment.