OpenAI led a coalition of researchers that discovered critical security vulnerabilities in PHP, the open-source programming language that powers roughly 70% of the world's websites. Governments across Europe and Canada took their websites offline to install patches before attackers could exploit the flaws. The FBI warned that malicious actors could use these vulnerabilities to take control of sites, steal data, or install malware. OpenAI reported that AI tools now help identify hundreds of critical vulnerabilities per month, a significant jump from previous years.
Anthropic launched Claude Duet, a new AI agent aimed at professionals and small business owners. Built around Claude Opus 5.5, the agent sends daily briefings, monitors connected apps, and handles tasks like scheduling meetings, building dashboards, and creating forecasts. Claude Duet asks for user approval before performing sensitive actions and includes safeguards to protect strict-tier information such as credit card numbers. The product enters the market as a research preview.
A new report from IRONSCALES found that 77% of organizations experienced agentic AI-related security incidents in the past year. Of the 501 IT and cybersecurity professionals surveyed, 96% said their organizations now use AI agents. About 49% give agents access to email or messaging platforms, and 30% grant access to finance or payment systems. Attackers are increasingly using phishing and social engineering tactics to exploit these AI agents.
Former US Treasury secretary Robert Rubin urged the United States and China to cooperate on AI risks, drawing a parallel to Cold War-era nuclear arms control. Speaking at the Bund Summit in Shanghai, Rubin cited the Strategic Arms Limitation Talks as a model and said both countries should set aside distrust to address issues that could severely affect them. Separately, an opinion piece called for a Manhattan Project-style effort to prepare for biological risks posed by AI, arguing that uncertainty about harm is not a reason to delay preparation.
California's Little Hoover Commission reported strong progress on six of nine AI recommendations, including launching an AI app called Poppy for state employees and planning a public AI data center. However, the state has not funded a public computing facility as the commission recommended. The commission also called for strengthening worker voices and improving public education about AI. Meanwhile, Brown University's Division of Biology and Medicine announced plans to expand AI initiatives in research and create a dedicated leadership role for AI efforts, though federal funding uncertainty is pushing the institution toward more partnerships with foundations and industry.
Key Takeaways
- <ul><li>OpenAI led a coalition that discovered critical PHP vulnerabilities affecting 70% of websites, prompting governments in Europe and Canada to take sites offline for patches</li><li>AI tools now identify hundreds of critical vulnerabilities per month, according to OpenAI, far exceeding past years</li><li>Anthropic launched Claude Duet, an AI agent built on Claude Opus 5.5, designed for professionals and small business owners with daily briefings and app monitoring</li><li>77% of organizations reported AI agent security incidents in the past year, with 96% now using AI agents, per an IRONSCALES survey of 501 professionals</li><li>Robert Rubin urged US-China cooperation on AI risks at the Bund Summit in Shanghai, comparing it to Cold War arms control</li><li>An opinion piece calls for a Manhattan Project-style biodefense effort to address biological risks from AI</li><li>California launched an AI app called Poppy for state employees but has not funded a recommended public computing facility</li><li>Brown University's BioMed division plans to expand AI research and create a dedicated AI leadership role amid federal funding uncertainty</li><li>Hunter Engineering will showcase nearly two dozen products at the 2026 SEMA Show, including an AI-based product</li><li>The aviation parts trading industry is adopting AI for forecasting and pricing, though data quality remains a challenge</li></ul>
AI discovery of PHP flaws takes websites offline worldwide
AI tools helped find serious security flaws in PHP, a programming language that runs about 70% of websites. Governments and organizations across Europe and the U.S. took their websites offline to install fixes. The FBI warned that hackers could use the flaws to take control of websites and steal data. OpenAI led a coalition of researchers that found the vulnerabilities before attackers could exploit them.
AI-assisted researchers reveal major internet security flaws
Critical vulnerabilities in PHP, an open-source language behind sites like WordPress, were discovered with heavy AI assistance. Left unfixed, the flaws could let attackers take over websites, add malware or steal user data. Government websites in Europe and Canada were taken down for updates. OpenAI reported that AI tools now find hundreds of critical vulnerabilities per month, far more than in past years.
Opinion calls for Manhattan Project style biodefense effort
An opinion piece argues that the U.S. needs a large coordinated effort to prepare for biological risks posed by AI. The plan would focus on detecting and preventing misuse of biotechnology made easier by AI advances. The author says uncertainty about AI harm is not a reason to delay preparation. The effort would not involve secrecy but would require sustained commitment.
Brown BioMed Dean outlines AI plans and funding challenges
Dean Mukesh Jain spoke about the Division of Biology and Medicine's priorities at its annual State of BioMed address. The division plans to expand AI initiatives in research and coordinate a dedicated leadership role for AI efforts. Federal funding remains uncertain, so partnerships with foundations and industry are growing more important. A 2026 state report found Rhode Island needs at least 300 more primary care clinicians.
California advances AI use in government but gaps remain
A new report from California's Little Hoover Commission says the state has made strong progress on six of nine AI recommendations. Steps include launching an AI app called Poppy for state employees and planning a public AI data center. However, the state has not funded a public computing facility the commission recommended. The commission says California still needs to strengthen worker voices and improve public education about AI.
AI Transforms Aviation Parts Trading with Better Forecasting
The aviation parts trading industry is adopting AI to improve forecasting and pricing for USM trading. Experts say AI helps companies make more informed decisions. However, AI is not a replacement for human expertise. Data quality and transparency remain key challenges for the industry.
Hunter Engineering to Show AI Products at SEMA Show
Hunter Engineering Co. will showcase nearly two dozen product introductions at the 2026 SEMA Show in Las Vegas from Nov. 3 to 6. The company will preview an AI-based product and a new product series. Featured items include the ADASLink scan tool with Bosch support, the HawkEye XL alignment system, and the WalkAway wheel balancer system that cuts tire changeover time by up to 44%.
77% of Organizations Report AI Agent Security Incidents
A new IRONSCALES report found that 77% of organizations experienced agentic AI-related security incidents in the past year. The survey of 501 IT and cybersecurity professionals found that 96% of organizations now use AI agents. About 49% give agents access to email or messaging platforms, and 30% grant access to finance or payment systems. Attackers are using phishing and social engineering to exploit these AI agents.
Anthropic Launches Claude Duet AI Agent for Professionals
Anthropic announced Claude Duet, a new AI agent designed for professionals and small business owners. It launches as a research preview and is built around Claude Opus 5.5. The agent sends daily briefings, monitors connected apps, and handles tasks like scheduling meetings, building dashboards, and creating forecasts. Claude Duet asks for user approval before taking sensitive actions and includes safety features to protect strict-tier information like credit card numbers.
US and China Urged to Cooperate on AI Risks Like Cold War
Former US Treasury secretary Robert Rubin urged Washington and Beijing to cooperate on AI risks, drawing a parallel to Cold War-era nuclear arms control. Rubin cited the Strategic Arms Limitation Talks between the US and Soviet Union as a model. He spoke at the Bund Summit in Shanghai on Monday. Rubin said the two countries should set aside distrust to address issues that could severely affect both nations.
Sources
- Some websites go dark as operators address an AI-detected flaw in a web programming language
- AI helped find a critical bug affecting huge swathes of the internet
- AI capabilities mean America needs a Manhattan Project for biodefense
- Dean Jain talks funding, AI initiatives, community engagement at annual ‘State of BioMed’ address
- Report: Progress on AI in state government but more to be done
- MRO Memo: Part-ificial Intelligence? AI Comes To USM Trading
- Hunter to Showcase New Technology at SEMA and AAPEX
- New Report From IRONSCALES Reveals 77% of Organizations Experienced Agentic AI-Related Security Incidents in the Past Year
- Anthropic takes on Muse with Claude Duet, its new AI agent
- US, China should learn from Cold War-era arms control on AI: ex-US official
Comments
Please log in to post a comment.