Research published in Nature and The Lancet Digital Health raises concerns that pharmaceutical advertising can influence AI treatment recommendations. In tests of 12 models from Google, OpenAI, Anthropic and other companies, models selected an advertised drug in 47.6% of responses when an ad was present, versus 35% without one. Across 258,559 responses to 33 physician-written cases, ads could affect choices even when they were unrelated to the question; some recommendations were not evidence-based. Researchers caution that the answers could still appear medically reasonable. The studies did not examine effects on prescribing, patient choices, costs or health outcomes.
AI adoption is also accelerating across workplaces, though oversight and employee concerns remain. A Rootstock Software survey found 41% of manufacturing respondents expect AI to improve productivity, while 61% worry it could limit career opportunities and 44% cite accuracy, bias and unpredictable results as a leading barrier. Still, 77% of organizations plan to increase AI resources in the next 12 to 18 months. Among investment firms, a May 2026 survey found 87% have deployed AI or established use cases, mostly for operations and analysis; employee readiness is the main obstacle, cited by 61%. Over three to five years, 42% plan to embed AI in investment decisions.
Milestone CEO Liad Elidan says companies need to measure whether AI coding tools improve products and justify their costs. His firm tracks AI use, spending, engineering output, code quality and how long code remains in production at enterprises with at least 200 engineers. Elidan gave the example of a senior engineer using five agents potentially costing 25% more while producing three times the output, while warning that the value is unclear without measurement and that larger code volumes can strain human reviews.
Several organizations are setting out practical approaches to AI. Google has introduced Agent Substrate, an open-source project donated to the Cloud Native Computing Foundation. Built on Kubernetes and using gVisor, it isolates agents running code or browsing the web, supports policy-controlled credential access and can save idle-agent snapshots to free resources. Google already uses it in Gemini Managed Agents and Agent Compute Engine. Fish Audio, meanwhile, released Live 3, a full-duplex voice model that can listen while speaking, recognize interruptions and make short reactions in as little as 186 milliseconds. It is available through the Fish Agent Platform and APIs for $0.035 per minute through Dec. 31, 2026; phone calls cost an additional $0.01 per minute.
Avanade plans to acquire UK automation firm Robiquity to combine its global reach with Robiquity’s Microsoft expertise and help UK mid-market and enterprise clients use AI and automation. Avanade CEO Chris Howarth praised Robiquity’s team, client focus and learning culture. In the U.S. Air Force, plans call for AI tools across 15 missions, including intelligence and nuclear operations, with an assigned civilian or officer overseeing each area. Three training levels are due by the end of November, from general AI principles to workflow building and developer training; participation will be recorded in service members’ training records.
AI safety and governance also featured in public and emergency-preparedness discussions. University of Wisconsin chancellor finalist Dr. Rashid Bashir said at a Nov. 9, 2026, community event that AI should be used responsibly, with human involvement and in ways suited to people’s expertise and age. He also stressed campus safety, productive disagreement and timely support for survivors of interpersonal violence. Separately, more than 115 Illinois hospitals and health systems joined a Nov. 4 exercise simulating an AI-driven cyberattack, including deepfakes and impersonations of leaders, that disrupted care and communications while systems appeared to work. The Illinois Health and Hospital Association called it the state’s largest AI-focused hospital preparedness exercise.
Key Takeaways
- Clinical tests found models chose an advertised drug in 47.6% of responses with an ad, compared with 35% without one; the studies did not measure real-world prescribing or health outcomes.
- Manufacturing workers see potential productivity gains, but 61% worry AI could constrain careers; 77% of surveyed organizations plan to increase AI resources in the next 12 to 18 months.
- In a survey of investment firms, 87% had deployed AI or set up use cases, but adoption readiness was the leading barrier; 42% plan to embed AI in investment decisions over the next three to five years.
- Milestone CEO Liad Elidan urges companies to track AI coding costs, output and code quality rather than assume increased code volume creates value.
- Google’s open-source Agent Substrate uses Kubernetes and gVisor to isolate AI agents and is already used in Gemini Managed Agents and Agent Compute Engine.
- Fish Audio Live 3 supports full-duplex voice interaction, with short reactions as fast as 186 milliseconds; its listed price is $0.035 per minute through Dec. 31, 2026.
- Avanade plans to acquire Robiquity, bringing together Avanade’s global reach and Robiquity’s Microsoft expertise; CEO Chris Howarth praised the company’s client focus and culture.
- The Air Force plans AI tools for 15 missions and three training levels by the end of November, with course participation added to permanent training records.
- More than 115 Illinois hospitals and health systems participated in a Nov. 4 exercise simulating AI-enabled cyber disruption, deepfakes and impersonation attacks.
- UW chancellor finalist Rashid Bashir called for responsible AI use with human involvement and emphasized campus safety and timely survivor support.
Drug ads can influence AI treatment recommendations
Researchers at the Icahn School of Medicine at Mount Sinai found that commercial ads can affect AI treatment recommendations, even when the ads are not related to the health question. In a study using 1,000 clinical questions, AI systems more often recommended treatments that appeared in the dataset, including some treatments that were not based on evidence. The study was conducted with researchers from the University of California, San Francisco, and the University of Pennsylvania, and received funding from the National Institutes of Health. Published in Nature, the study calls for AI systems that limit advertising’s influence and for health providers to watch for possible bias.
Drug ads shift AI treatment choices in clinical tests
A study published Nov. 11 in The Lancet Digital Health found that pharmaceutical ads can change treatment choices made by AI models. Researchers at the Icahn School of Medicine at Mount Sinai tested 12 models from companies including Google, OpenAI and Anthropic on 33 physician-written cases. The models chose the advertised drug in 47.6% of responses when an ad appeared, compared with 35% without an ad, across 258,559 responses. The models rarely named the ad, and researchers said the answers could still seem medically reasonable; the study did not test effects on prescriptions, patient choices, costs or health outcomes.
AI use grows in manufacturing as workers voice concerns
A Rootstock Software survey found that AI use is growing in manufacturing, but workers still have concerns. More than a quarter of respondents said they are seeing results from AI, while 18% were in pilot programs and 13% had operational AI with no results. Although 41% expect AI to improve their productivity, 61% worry it could limit their career opportunities; concerns about accuracy, bias and unpredictable results were the top barrier, cited by 44%. Still, 77% said their organizations plan to increase AI budgets or resources over the next 12 to 18 months, and 88% saw value in a unified AI agent across business applications.
UW chancellor finalist Rashid Bashir discusses AI and campus safety
Dr. Rashid Bashir, the first of three finalists for University of Wisconsin chancellor to speak, answered community questions at Memorial Union on Nov. 9, 2026. He said AI should be developed and used responsibly, with a human involved, and should support people in ways suited to their expertise and age. On campus expression, Bashir said students should feel safe sharing strong and varied views, and he connected this to a community that supports productive disagreement. He also said universities have a responsibility to protect students from interpersonal violence, including sexual assault and hazing, and that survivor support programs need clear, timely responses.
Fish Audio Live 3 brings 186 millisecond reactions to voice agents
Fish Audio released Live 3, a full-duplex voice model for real-time voice agents that can listen while speaking. The company says the model learned conversational timing from real human conversations, helping it respond to short acknowledgments, recognize interruptions and wait for corrections. It can decide to make short reactions in as little as 186 milliseconds, while preparing a full answer in parallel; Fish Audio has not published a latency distribution. Live 3 is available through the Fish Agent Platform and related APIs, and its listed price is $0.035 per minute through Dec. 31, 2026, down from $0.07; phone calls cost an extra $0.01 per minute.
Investment firms adopt AI, but few use it for key decisions
A May 2026 survey of more than 300 senior decision makers found that 87% of investment firms have deployed AI or set up use cases. Most use it for operational gains and analysis, while only a few use AI for investment decisions or risk management. The leading barrier is employee readiness and adoption, cited by 61%, followed by shortages in AI engineering and technical skills. Many firms also lack complete data systems and connected workflows. Over the next three to five years, 45% plan to prioritize modernizing legacy systems, 42% plan to embed AI in investment decisions, and 40% aim to improve the speed and confidence of capital allocation.
Avanade plans to acquire UK automation firm Robiquity
Avanade plans to acquire Robiquity, combining its global reach with Robiquity’s Microsoft expertise. The deal is intended to help mid-market and enterprise clients across the UK use AI and automation to improve productivity, streamline operations and find new growth opportunities. Avanade CEO Chris Howarth praised Robiquity’s team, focus on client outcomes and culture of continuous learning. The article was published on November 11, 2026.
Milestone CEO warns companies lack oversight of AI coding costs and quality
Liad Elidan, co-founder and CEO of Israeli startup Milestone, said many companies are adopting AI coding tools without checking whether they improve products or justify their cost. Milestone works with enterprises that have at least 200 engineers and tracks AI use, cost, engineering output, code quality and how long code stays in production. Elidan warned that powerful models can drive up spending and that larger volumes of code can overwhelm human reviews. He said a senior engineer using five agents might cost 25% more while producing three times the output, but the value is unclear if costs and results are not measured. He predicted that by 2030 one person could direct AI agents across marketing, design and coding.
Air Force plans three levels of AI training as it prepares new tools
The Department of the Air Force plans to introduce AI tools for 15 missions, from intelligence to nuclear operations, and assign a civilian or officer to oversee each area. Chief data and AI officer Susan Davenport said the department will launch three levels of AI training by the end of November. The 100-level course will cover general AI principles, the 200-level will teach people to build AI-powered workflows, and the 300-level will target developers. Course participation will go into service members’ permanent training records. The department also plans quarterly showcases to share progress and connect mission teams with researchers and private companies; the third is planned for December or January.
Illinois hospitals test response to an AI-driven cyberattack
More than 115 Illinois hospitals and health systems took part in the “Broken Trust: AI Takes Control” emergency exercise on November 4. Led by the Illinois Health and Hospital Association, the virtual exercise simulated an AI-driven attack that disrupted patient care, communications and hospital operations while leaving systems apparently working. The scenario included AI-generated impersonations of leaders, communication failures and deepfake videos, requiring teams to check whether information could be trusted. OSF HealthCare participated, and director Sherry Myers said it works with emergency management, IT and clinical teams to maintain essential services when technology is unavailable. The association said it was the largest AI-focused hospital emergency preparedness exercise ever conducted in Illinois.
Google opens secure sandboxing project for AI agents
Google introduced Agent Substrate, an open-source project built on Kubernetes and donated to the Cloud Native Computing Foundation. It uses gVisor to isolate AI agents as they run code or browse the web, helping protect enterprise tools and credentials. The system sends credentials directly when needed so external policies can control agent actions. It can also save snapshots of idle agents to free up CPU and memory. Google already uses Agent Substrate in Gemini Managed Agents and Agent Compute Engine.
Sources
- Drug ads can subtly sway AI medical treatment recommendations
- AI chatbot treatment recommendations are swayed by drug ads
- AI Adoption Grows in Manufacturing, but Worker Concerns Persist
- UW Chancellor finalist on AI usage, freedom of expression and interpersonal violence
- Fish Audio Releases Live 3, a Full-Duplex Voice Model That Reacts in 186 ms
- Despite Broad AI Adoption in Investment Management, Applications Remains Narrow - Fintech Schweiz Digital Finance News
- Avanade to Acquire Robiquity, Bringing AI-First Automation to More UK Businesses
- AI is making engineers faster, but nobody is checking the bill or the quality, says AI Israeli CEO
- Air Force previews AI training as it prepares to deploy new tools
- OSF joins statewide AI cyberattack exercise to strengthen hospital emergency response
- Google releases open-source agent substrate for secure AI sandboxing
Comments
Please log in to post a comment.