Google OpenAI and Anthropic AI Models Favor Advertised Drugs

Key Takeaways

• SentinelOne says Singularity Agent Detection and Response can identify AI agents, enforce approved-tool rules and help stop agents or revoke access when their actions diverge from a user’s request.
• Reviews of AI agent skills, tools and MCP servers should examine authorship, permissions, system access and enabled actions, including risks that are not conventional software vulnerabilities.
• Mount Sinai researchers found that Google, OpenAI and Anthropic models selected an advertised drug in 47.6% of responses with ads, compared with 35% without ads; the study did not measure prescribing or health outcomes.
• HUD plans a data and AI division and broader reorganization through fiscal year 2027; 41% of employees who left in September 2025 had served more than 20 years.
• An AI workplace training program launched November 7, 2026, with over 4,000 professionals from 271 institutions in 13 countries; participants will complete workplace projects and train colleagues.
• Box cofounder Aaron Levie discussed AI’s potential and its use in Box and customer workflows; Carbonez says AI reduces decarbonization planning from months to under two hours.
• At Seoul’s November 13, 2026, drone standards assembly, participants discussed hydrogen, AI and cybersecurity; South Korea announced investment plans focused on hydrogen fuel cells and AI.
• AI-oriented data centers require more power and computing capacity, raising local concerns about electricity and water even as supporters cite construction jobs and tax revenue.
• A November 12, 2026, analysis says Europe’s Digital Networks Act should clarify telecom rules and encourage investment in fibre, optical transport and edge computing for AI infrastructure.
• León Castellanos-Jankiewicz urges a binding international treaty on autonomous weapons, with enforceable duties for states and liability for companies building unpredictable systems.

SentinelOne outlines detection and response for AI agents

Corde describes SentinelOne’s Singularity Agent Detection and Response as a way to find active AI agents, set rules for approved tools, and detect unwanted behavior. Security teams may be able to stop an agent or remove its access to tools. The approach aims to check whether an agent’s actions match the user’s original request, not just whether the agent appears malicious. Corde also discusses Singularity ASOC, which uses agents to help speed up security operations workflows.

Review AI agent tools before giving them access

AI agents can gain access to data, systems, and commands through skills, tools, and MCP servers, so teams should review these components before deployment. A component may have excessive permissions or lead to unintended actions, even if it has no obvious software vulnerability. Reviews should examine who made each component, what access it requests, what systems it can reach, and what actions it enables. The article says assessments should explain the behavior and the conditions that could make a concern exploitable.

Box cofounder Aaron Levie shares his enthusiasm for AI

Box cofounder Aaron Levie discussed why he writes about AI and how quickly he expects the technology to reach its potential. He also described how Box and its customers use AI in their workflows. Levie, who cofounded Box in 2005, has long spoken about technology’s effects on productivity. The interview focused on his views about AI and its growing role in work.

Seoul drone assembly explores hydrogen, AI, and cybersecurity

The International Drone Standards Assembly took place in Seoul, South Korea, on November 13, 2026. Organized by the nonprofit IDSA, it brought together government, industry, and academic representatives to discuss drone standards. Topics included hydrogen fuel cells, artificial intelligence, and cybersecurity for next-generation drones. The IDSA plans to hold regular assemblies, with the next scheduled for 2027. South Korea’s government also announced plans to invest in drone technology, focusing on hydrogen fuel cells and AI.

Study finds drug ads can sway AI treatment recommendations

Researchers from the Icahn School of Medicine at Mount Sinai tested 12 AI models from Google, OpenAI, and Anthropic across 33 clinical scenarios. In 258,559 responses, models chose an advertised drug 47.6% of the time when an ad appeared, compared with 35% without one. Ads had the greatest effect when more than one treatment was medically reasonable, though models generally resisted promotion of unsupported or inferior products. The researchers did not determine whether the changes affected prescriptions or health outcomes, and called for greater transparency and testing.

HUD plans new data and AI division in major reorganization

The Department of Housing and Urban Development plans to create a data and artificial intelligence division as part of a larger reorganization of its headquarters and field operations. HUD says the new unit will help give data and AI the staff and attention they need, while supporting industry practices in product management, development, security and operations. In September 2025, more than half of the employees who left HUD had worked there for at least 10 years, and 41 percent had served more than 20 years. The reorganization is expected to start about 90 days after the notice is published and continue through fiscal year 2027. HUD is also preparing technology purchases related to AI, innovation, compliance and financial systems modernization.

Experts call for new rules on autonomous weapons and AI

León Castellanos-Jankiewicz argues that current international law does not fully address weapons that use artificial intelligence and can act with limited human control. He says people have handed some thinking and decisions in warfare to machines, creating gaps in accountability. AI-guided drones may attack targets based on learned categories without operators seeing the specific target, making responsibility harder to establish. The article urges officials meeting in Switzerland to begin talks on a binding treaty. It says new rules should set enforceable duties for states using unpredictable weapons and create liability for the companies that build them.

AI training program aims for real workplace results

The Academic and Grant Writing Enclave says AI training should lead to measurable workplace improvements, not just certificates. Programme Coordinator Dr Wadzani Dauda Palnam made the call during an international program that began on November 7, 2026, with more than 4,000 professionals from 271 institutions across 13 countries. Six virtual sessions will be followed by projects at participating universities, organizations and government ministries. Participants are expected to use AI to solve workplace problems and teach the skills to 30 to 50 colleagues. Experts leading sessions include Professor Yuko Harayama, Professor Catholijn M. Jonker and Professor Virginia Dignum, alongside AGE Board of Studies Chairman Professor Charles Okechukwu Esimone.

Europe’s Digital Networks Act must support AI infrastructure

Europe’s proposed Digital Networks Act could shape the connectivity infrastructure needed for artificial intelligence, but the article warns that unclear rules could add costs and slow deployment. It calls for a clear separation between telecom networks and digital services such as enterprise applications, as well as simpler telecom rules. The article says AI depends on an integrated system of connectivity, computing and AI, including secure data-center links and high-capacity wireless networks. It argues that Europe needs stronger investment incentives for fibre, optical transport and edge computing, with funding discussions tied to the future multiannual framework and European Competitiveness Fund. The article is dated November 12, 2026, and discusses the European Commission proposal and draft report by Parliament Rapporteur Michał Kobosko.

AI demand fuels bigger data centers and local debate

Data centers are growing in size and number as demand for cloud storage and artificial intelligence increases, including in the Tri-State area. They store and process data used for everyday activities such as email, online banking and social media, said Srishti Srivastava, an Associate Professor of Computer Science at the University of Southern Indiana. Many newer centers are designed for AI, which needs more power and larger computer systems. Supporters say data centers can create construction jobs and tax revenue, while concerns include their use of electricity and water. Reliable power has become a major challenge, leading some operators to build power sources on-site or seek new locations.

Carbonez uses AI to speed up business decarbonization

Carbonez is changing how businesses plan to reduce their carbon emissions. The process once took months, but Carbonez can now complete it in less than two hours. Erdem Celek explains how the company uses AI to speed up decarbonization planning.

Sources

• AI Agents Need Their Own Detection and Response
• Before You Let AI Agents Act, Know What They’re Running
• Box’s Aaron Levie is having fun with AI, and thinks you should, too
• International Drone Standards Assembly Held in Seoul
• Drug ads can influence AI treatment recommendations, study finds
• HUD OCIO Plans Data and Artificial Intelligence Division
• AI Is Killing Autonomously. Time for New Rules.
• ‘AI training must go beyond certificates, deliver workplace results’
• The Digital Networks Act: Building the connectivity infrastructure Europe needs for AI
• Data centers expand in size and number as AI drives demand
• Can AI help solve the climate crisis faster?