OpenAI's decision to halt GPT-6.1 Astra and increase scrutiny of agentic models has drawn fresh attention to AI security across the tech industry. NVIDIA responded by launching the Open Agent Safety Platform on September 28, 2026, with support from companies like OpenAI, Anthropic, and Meta. The push reflects growing concern that model-level protections alone are no longer sufficient as AI agents become more capable and autonomous.
Chipmakers and infrastructure suppliers are feeling both the opportunity and pressure. TSMC, Samsung Electronics, and SK hynix are central to secure computing and AI server hardware, though SK hynix faces risk from US export controls on China. Nanya Technology and CXMT also supply DRAM and DDR4/DR5 chips for data centers, while Intel provides CPUs and foundry services. On the networking side, Arista Networks, Lumentum, and Vertiv supply critical gear, with Vertiv carrying a US$15 billion backlog.
Higher interest rates from the Federal Reserve add financial strain on these AI infrastructure stocks, even as demand for data center capacity keeps climbing. Samsung and partners recently committed US$1 billion to an AI data center builder, signaling where capital is flowing despite economic uncertainty.
On the policy front, House Speaker Mike Johnson said he supports voluntary AI regulation, emphasizing external auditing and transparency. The White House planned a meeting with leaders from Google, Amazon, and Microsoft to discuss a framework, aiming to balance innovation with responsible use.
Universities are also expanding AI adoption. The University of Georgia launched AI @ UGA, offering free generative AI courses and tools like ChatGPT Edu, Gemini Pro for Education, and Microsoft 365 Copilot. Georgia Tech uses AI to speed up transfer credit evaluations and handle over 66,000 emails and 10,000 phone calls through AI agents. Meanwhile, Memphis is emerging as an AI research hub, with the University of Tennessee Health Sciences earning R1 status and xAI and Google expanding nearby operations.
Alibaba Qwen released Qwen-Audio-3.1-Realtime, a full-duplex voice model that can think, act, and decide when to speak. Available as a managed API on QwenCloud, prices dropped roughly 85 percent for Realtime and 70 percent for TTS. Cloudflare tested its Web Application Firewall against frontier AI models acting as hackers across 45 scenarios, finding that LLMs mutated attack payloads faster than human hackers.
Key Takeaways
- <ul><li>OpenAI halted GPT-6.1 Astra amid heightened AI security scrutiny, prompting NVIDIA to launch the Open Agent Safety Platform on September 28, 2026.</li><li>NVIDIA's Open Agent Safety Platform has backing from OpenAI, Anthropic, and Meta, signaling industry-wide concern over agentic AI security.</li><li>TSMC, Samsung, and SK hynix are key suppliers for secure AI computing, though SK hynix faces risk from US export controls on China.</li><li>Nanya Technology expects 30% growth in bit shipments for AI data centers, while Intel holds a US$64.6 billion market cap supplying CPUs and foundry services.</li><li>Vertiv has a US$15 billion backlog for AI data center power and cooling systems, while Samsung and partners committed US$1 billion to an AI data center builder.</li><li>House Speaker Mike Johnson supports voluntary AI regulation with external auditing; the White House planned a meeting with Google, Amazon, and Microsoft.</li><li>The University of Georgia expanded AI tools including ChatGPT Edu, Gemini Pro for Education, and Microsoft 365 Copilot for students and faculty.</li><li>Georgia Tech uses AI agents to handle over 66,000 emails and 10,000 phone calls annually and speeds transfer credit evaluation for 1,400 transfer students.</li><li>Alibaba Qwen launched Qwen-Audio-3.1-Realtime, a full-duplex voice model with prices down 85% for Realtime and 70% for TTS on QwenCloud.</li><li>Cloudflare tested its WAF against frontier AI models across 45 scenarios, finding LLMs mutated attacks faster than human hackers.</li></ul>
AI Safety Spotlight Drives Interest in TSMC Samsung and SK hynix
OpenAI's halt of GPT-6.1 Astra and scrutiny of agentic models has raised focus on AI security. TSMC fabricates chips for secure computing environments. Samsung Electronics supplies memory and logic chips for AI servers with revenue around ₩1,722.7 trillion in market cap. SK hynix provides DRAM and NAND for data centers but faces risk from US export controls on China. All three companies are exposed to the growing importance of secure cloud computing.
Three AI Infrastructure Stocks Face Higher Rates and Data Center Growth
Rising long-term yields and higher interest rates from the Federal Reserve create pressure on stock prices. Nanya Technology makes DRAM memory for AI data centers with market value around NT$1.7 trillion and expects 30% growth in bit shipments. CXMT produces DDR4 and DDR5 chips for AI servers with a market value of about CN¥3,652.2 billion. Intel supplies CPUs and foundry services with a market cap of US$64.6 billion. All three depend on stable infrastructure demand amid financial uncertainty.
Arista Networks and Two AI Infrastructure Stocks Worth Watching
Samsung and partners committed US$1 billion to an AI data center builder showing where investment is flowing. Arista Networks builds networking gear for AI data centers with a market value of about US$260.5 billion. Lumentum makes optical components for data center bandwidth with roughly US$84.5 billion in market value. Vertiv designs power and cooling systems for AI data centers with a market value near US$97.5 billion and a US$15 billion backlog. These companies provide essential infrastructure behind AI models.
University of Georgia Expands AI Tools and Resources for Campus
The University of Georgia launched AI @ UGA, a central hub for online courses and AI resources. Students can take a free course on generative AI. Faculty receive guides from the Center for Teaching and Learning on using AI in instruction. The university started a student pilot program in spring 2026 and expanded it in fall 2026 with ChatGPT Edu, Gemini Pro for Education and Microsoft 365 Copilot. UGA also offers a new Certificate in Artificial Intelligence and funds interdisciplinary AI research.
Georgia Tech Uses AI to Improve College Admissions and Financial Aid
Georgia Tech's Enrollment Management division is using AI to help students through the application process. AI tools speed up transfer credit evaluation for over 1,400 transfer students each year. Machine learning helps admission counselors review high school transcripts. The university is adding AI agents for 24/7 customer service to handle more than 66,000 emails and 10,000 phone calls received last year. AI also helps identify students who need to complete financial aid applications so they do not miss scholarship opportunities.
NVIDIA launches Open Agent Safety Platform for AI security
NVIDIA announced the Open Agent Safety Platform on September 28, 2026 to address growing AI security concerns. Companies like OpenAI, Anthropic, Meta, and Google have reported security breaches involving their AI models. Justin Boitano, NVIDIA's VP of Enterprise AI, said model-level protections are no longer enough. NVIDIA is a major supplier of AI infrastructure with a market cap of about $5.52 billion.
Artificial intelligence and credit cards who pays the debt
This opinion piece explores the connection between artificial intelligence and credit card debt. The article was published in connection with scheduled maintenance on the News Gazette Digital platform from March 15 to March 16, 2025. The author Paul Selvin is a professor of physics and biophysics at the University of Illinois.
Software investing faces new challenges from AI and slower growth
A Bain report shows software sector revenue growth has dropped from about 20% annually to half that rate. Net revenue retention has fallen about 9 points since 2021. AI is changing software economics and threatening the SaaS value proposition. Private equity firms must adapt their investment strategies to account for AI-native challengers and aging portfolios that underperform pre-Covid averages.
Computer science offers lessons for controlling AI agents
An article discusses how AI agents from OpenAI, Anthropic, and Google have been involved in hacking incidents in 2026. The author draws lessons from the 1983 movie War Games and computer science research. The article recommends that organizations audit their security systems and design AI agents to authenticate themselves. As of September 2026, AI attacks have targeted nonvital government sites and smaller companies.
Cloudflare tests WAF against frontier AI models
Cloudflare tested its own Web Application Firewall using frontier AI models acting as hackers. The team ran 45 scenarios covering six attack categories against an authorized customer staging environment. The LLMs were able to iterate and mutate attack payloads faster than human hackers. The tester was built in Python and both model calls had no access to internal WAF information.
Mike Johnson pushes voluntary AI regulation to ease tech industry concerns
House Speaker Mike Johnson said he supports voluntary regulation of artificial intelligence. He wants external auditing to keep AI safe and transparent. The White House planned a meeting with leaders from Google, Amazon, and Microsoft on Friday. Johnson said the goal is to protect innovation while making sure AI is used responsibly. A bill is being worked on to create this voluntary framework.
Memphis becomes a hub for AI research through its universities
The University of Tennessee Health Sciences in Memphis earned R1 research status in 2025. This puts Memphis among only nine U.S. cities with two top research schools. AI company xAI moved to the area and Google is building a data center nearby. Researchers at UT Health Sciences are already using AI in drug discovery and patient care. Professors say AI is helping speed up research and improve medical treatments.
Alibaba Qwen launches new real-time voice AI model
Alibaba Qwen released Qwen-Audio-3.1-Realtime, a full-duplex voice model for voice agents. The model can think, act, and decide when to speak. It is available as a managed API on QwenCloud. Prices dropped by about 85 percent for Realtime and 70 percent for TTS. The model supports function calling, web search, reasoning, and audio input and output.
Sources
- AI Safety News Puts TSMC Stock And Secure Cloud Plays In Focus
- 3 AI Infrastructure Stocks Facing Higher Rates And Rising Data Center Demand
- Arista Networks And 2 Other AI Infrastructure Stocks To Own
- UGA expands AI resources for campus community
- How AI Supports Undergraduate Admission, Enrollment
- NVIDIA Introduces Open Agent Safety Platform Amid AI Security Concerns
- My Turn | Artificial intelligence and credit cards: Who pays the debt?
- Software Investing in the Age of AI and Slower Growth
- The ‘War Games’ problem: Computer science has long understood what it takes to keep AI under control
- We tested our own WAF with frontier AI models. Here’s what we found
- Mike Johnson backs voluntary regulation to ‘calm the nerves’ on AI safety - Live Updates
- Where Academic Healthcare Meets AI
- Alibaba Qwen Releases Qwen-Audio-3.1-Realtime: A Full-Duplex Voice Model Trained to Think, Act, and Decide When to Speak
Comments
Please log in to post a comment.