OpenAI's automated security pipeline found 948 working local privilege escalation exploits in the Linux kernel during a single run in mid-October 2026. The system used scout agents, analyst agents, and exploit development agents working in stages to narrow down 40 million lines of kernel code to 5,034 audit targets. The total cost was $1,213,430 in API credits, roughly $1,279 per exploit, which OpenAI says could be worth $66.3 million compared to historical contest bounties.
Every exploit and patch was reviewed by human researchers before being reported to the kernel community. The pipeline confirmed 1,724 bugs out of 2,308 candidates, with 74.7 percent surviving dynamic validation. OpenAI submitted hundreds of reports and patches, and at least 10 new Linux exploitation techniques were discovered. Experts note that many of the found flaws were already patched within 10 days, showing that defenders are also improving.
In the banking sector, experts are raising alarms about AI agents making financial decisions without proper safeguards. Jake Williams and T.J. Marlin of Guardrail Technologies warn that mistakes could cost consumers large amounts of money and call for clear limits, stop mechanisms, and records on AI agent actions. Comptroller of the Currency Jonathan Gould says banks are working to defend against rogue AI agents, while consumers must decide how much control they are willing to give AI with their money.
onsemi reported a 175% surge in AI sales for its third quarter, beating expectations. The company's power-delivery technology for AI processors drove an 11% revenue growth in Q3. onsemi is acquiring Ferric to strengthen its position in the AI market, and analysts are predicting a stock recovery for the company next year.
Healthcare continues to struggle with governing AI tools it has not fully identified. Nearly 90% of healthcare leaders say security staff review AI-generated code, but only half have a written policy on AI coding assistants. More than half of leaders report that AI-generated code has led to hardcoded credentials, and about half report exposure of patient data or insecure data storage. Paubox advises setting rules for test work to protect health information under HIPAA.
CrowdStrike has uncovered evidence that VAPOR PANDA, a China-aligned hacking group, used a multi-agent AI framework called ARTEX to coordinate attacks across at least 14 countries in just 11 days. The framework directed planner and worker agents to conduct reconnaissance, steal credentials, move laterally, and deploy web shells. In one intrusion, planning agents consumed about 1.4 billion input tokens. Targets included government entities, pharmaceutical companies, semiconductor firms, and defense organizations.
Microsoft announced five key updates to its Microsoft 365 Education portfolio, including a no-cost student option and a new A7 tier with advanced security and AI. Microsoft Classroom brings AI-powered learning tools and the Study and Learn Agent for K-12 and higher education. A2-A5 offerings provide discounts over 80% compared with commercial pricing, and A7 offers up to 60% off. Most changes will take effect at future contract renewals starting May 1, 2027.
AMD has agreed to acquire Toronto AI chip startup Taalas for about $2.9 billion. The deal is expected to close by the end of 2026, pending regulatory approval. Taalas builds AI-inference chips directly into silicon and was founded by former AMD executives. The technology will complement AMD's Instinct MI400 Series and Helios rack-scale systems, and the Taalas team will join AMD's Canadian workforce with no planned layoffs.
Mariner Wealth Advisors says its AI agent system named Hugo uncovered $22 billion in client assets that advisors had missed, including forgotten 401(k) plans from former employers. The company currently serves 70,000 households and aims to reach 500,000 by the end of 2028. Junior staff are advancing to advisor roles faster because of Hugo, and 80 of the company's 200 client associates have raised their hands to become advisors.
VMware is expanding its Cloud Foundation with AI Infrastructure Quickstart, an automated tool that simplifies setting up private AI environments. A new multitenant model sharing feature lets multiple teams use shared AI models while keeping separate security boundaries. VMware also introduced vSphere Express, a lower-cost platform for environments with up to 128 processor cores. More than 2,600 customers had deployed VCF 9 as of September, covering over 24 million allocated cores.
Key Takeaways
- <ul><li>OpenAI's AI pipeline found 948 Linux kernel exploits in a single run, costing $1.2 million in API credits</li><li>OpenAI says the $1.2 million campaign could be worth $66.3 million compared to historical contest bounties</li><li>Human researchers reviewed every exploit and patch before reporting to the kernel community</li><li>VAPOR PANDA used multi-agent AI to attack targets across 14 countries in 11 days</li><li>Microsoft 365 Education gets new AI tools, no-cost student option, and discounted tiers</li><li>AMD to acquire Taalas for $2.9 billion to strengthen AI inference chip capabilities</li><li>onsemi reports 175% jump in AI sales driven by power-delivery technology for AI processors</li><li>Mariner's Hugo AI agent uncovered $22 billion in hidden client assets</li><li>Nearly 90% of healthcare leaders review AI-generated code, but only half have written policies</li><li>VMware launches AI Infrastructure Quickstart to simplify private AI deployment</li></ul>
OpenAI AI Pipeline Found 948 Linux Root Exploits in One Run
OpenAI used an AI pipeline to find 948 local privilege escalation exploits in the Linux kernel during a single mid-October 2026 run. The pipeline cost $1.2 million in API credits, about $1,279 per exploit. The system reviewed 2,308 potential bugs and confirmed 1,724 of them. Human researchers checked every exploit and patch before anything was reported.
AI Agents Exploited 948 Linux Kernel Flaws for OpenAI Study
OpenAI reported that AI agents built working root exploits for 948 unique Linux kernel vulnerabilities in one run. The mid-October scan targeted the latest kernel master branch code. The agents discovered at least 10 new Linux exploitation techniques. The total cost was $1,213,430 in API credits. OpenAI submitted hundreds of reports and patches to the kernel community.
OpenAI Made Local Root Finding Cheap at Scale
OpenAI's pipeline showed that finding local root vulnerabilities can now be done cheaply and at scale. Scout agents narrowed 40 million lines of kernel code down to 5,034 audit targets. The full run cost $1,213,430, or $1,279 per local privilege escalation exploit. Experts note that defenders are also improving, and many of the found flaws were already patched within 10 days.
OpenAI Cyber Model Discovered 948 Linux Kernel Exploits
OpenAI's automated kernel research pipeline found 948 working local privilege escalation exploits in mid-October 2026. The pipeline used scout agents, analyst agents, and exploit development agents in stages. Only 74.7 percent of candidate bugs survived dynamic validation. The campaign cost $1,213,430, which OpenAI says could be worth $66.3 million compared to historical contest bounties.
AI Agents in Banking Bring Benefits and New Risks
Experts warn that AI agents making financial decisions could lead to costly mistakes without proper safeguards. Jake Williams and T.J. Marlin stress the need for limits, stop mechanisms, and records on AI agent actions. Jonathan Gould says banks are using AI to defend against rogue AI agents. Consumers must decide how much control they are willing to give AI with their money.
AI agents in banking bring speed but new risks to consumers
Experts warn that AI agents used in banking need clear limits and oversight. These agents can make decisions and move money on behalf of users. Jake Williams and T.J. Marlin of Guardrail Technologies say mistakes could cost consumers large amounts of money. The Comptroller of the Currency Jonathan Gould says banks are working to defend against rogue AI agents. Consumers must decide how much control they are willing to give AI.
onsemi sees 175% jump in AI sales, eyes stock recovery
onsemi reported a 175% surge in AI sales for its third quarter, beating expectations. The company's power-delivery technology for AI processors drove an 11% revenue growth in Q3. onsemi is acquiring Ferric to strengthen its position in the AI market. Analysts are predicting a stock recovery for onsemi next year. AI sales have become a key driver of the company's revenue growth.
Healthcare struggles to govern AI tools it has not found
Nearly 90% of healthcare leaders say security staff review AI-generated code, but only half have a written policy on AI coding assistants. Paubox found that code review only covers what is submitted and business associate agreements only happen if someone reports a connection. More than half of leaders say AI-generated code has led to hardcoded credentials. About half report exposure of patient data or insecure data storage. Paubox advises setting rules for test work to protect health information under HIPAA.
VAPOR PANDA hackers use multi-agent AI to attack governments and industries
CrowdStrike found that VAPOR PANDA, a China-aligned hacking group, used an AI framework called ARTEX to coordinate attacks across at least 14 countries in 11 days. ARTEX directed planner and worker agents to conduct reconnaissance, steal credentials, move laterally, and deploy web shells. In one intrusion, planning agents consumed about 1.4 billion input tokens. The operator retained human oversight and directed agents in Chinese at key moments. Targets included government entities, pharmaceutical companies, semiconductor firms, and defense organizations. VAPOR PANDA also used BLOODALCHEMY backdoor and checked for missing CrowdStrike Falcon sensors to evade defenses.
Microsoft updates 365 Education with AI tools and new pricing
Microsoft announced five key updates to its Microsoft 365 Education portfolio. The changes include a no-cost student option, a new A7 tier with advanced security and AI, expanded faculty capabilities, scalable storage, and Microsoft Classroom for teaching and learning. Microsoft Classroom brings AI-powered learning tools and the Study and Learn Agent for K-12 and higher education. A2-A5 offerings provide discounts over 80% compared with commercial pricing, and A7 offers up to 60% off. Most changes will take effect at future contract renewals starting May 1, 2027.
AMD to acquire Toronto AI chip startup Taalas for $2.9 billion
AMD announced it will pay about US$2.9 billion for Taalas, a Toronto startup that builds AI-inference chips directly into silicon. The deal is expected to close by the end of 2026, pending regulatory approval. Taalas was founded by former AMD executives Ljubiša Bajić, Lejla Bajić, and Drago Ignjatovic. The technology will complement AMD's Instinct MI400 Series and Helios rack-scale systems. The Taalas team will join AMD's Canadian workforce with no planned layoffs.
Mariner's Hugo AI found $22 billion in hidden client assets
Mariner Wealth Advisors CEO Marty Bicknell said its AI agent system named Hugo uncovered $22 billion in client assets that advisors had missed. Mariner currently serves 70,000 households and aims to reach 500,000 by the end of 2028. Hugo helped identify overlooked assets including forgotten 401(k) plans from former employers. Junior staff at Mariner are advancing to advisor roles faster because of the Hugo AI workforce. The system has 200 client associates, 80 of whom have raised their hands to become advisors.
AI improves data handling in credit and trade finance underwriting
A new Gallagher Re Q3 2026 Global InsurTech Report says AI is closing the data gap in credit and trade finance underwriting. AI agents now process loan tapes that once took analysts two to four hours, reducing the task to minutes. Portfolio modeling that previously took a week now takes 10 to 20 minutes, according to Convergence, a Lloyd's-based credit insurer. Convergence grew about 140% in 2025 and processes over 500 credit insurance enquiries per month.
VMware launches simpler AI deployment and budget cloud option
Broadcom is expanding VMware Cloud Foundation with AI Infrastructure Quickstart, an automated tool that simplifies setting up private AI environments in minutes. A new multitenant model sharing feature lets multiple teams use shared AI models while keeping separate security boundaries. VMware also introduced vSphere Express, a lower-cost platform for environments with up to 128 processor cores. More than 2,600 customers had deployed VCF 9 as of September, covering over 24 million allocated cores. Some features are available now while others are planned for Broadcom's 2027 fiscal year.
Sources
- OpenAI AI Pipeline Generates 948 Linux Root Escalation Exploits in 30-Hour Run
- AI Agents Got Root on Linux via 948 Kernel Flaws in OpenAI Research
- OpenAI just made local root cheap at scale
- OpenAI cyber model found 948 Linux kernel exploits
- AI agents could make banking easier, but experts warn of new risks
- AI agents could make banking easier, but experts warn of new risks
- onsemi’s AI Sales Surge 175%: Is a Stock Recovery Next?
- Healthcare can't govern the AI tools it hasn't found yet
- VAPOR PANDA hackers use multi-agent AI to target government, pharmaceutical, semiconductor, defense organizations
- Future-ready education: Evolving Microsoft 365 Education
- AMD Taalas Acquisition: US$2.9B Toronto Deal [2026]
- Mariner’s Hugo AI uncovered $22B in client assets, says Marty Bicknell
- AI is closing the data gap in credit and trade finance underwriting, report finds
- VMware gets simpler AI deployment and low-cost version for basic private clouds
Comments
Please log in to post a comment.