AI governance and oversight remain uneven. A Society-Centered AI Governance Framework evaluates whether people can find out when AI influenced a decision, challenge it and secure a correction. Researchers applied it to North Carolina, interviewing nine AI Leadership Council members before and after the state’s 2026 AI roadmap and reviewing 49 governance documents from 19 states. North Carolina has visibility rules, but notice is an agency duty rather than an enforceable public right, and AI-specific challenges and corrections remain advisory. The roadmap assigns owners and deadlines, but those protections are still gaps.
Security researchers also warn that conventional checks may miss attacks on AI systems. Eclypsium, with support from NVIDIA, demonstrated that an attacker with privileged access could use a compromised data processing unit and weak memory isolation to alter model weights in memory and change answers during live inference—without changing stored files or triggering traditional integrity checks. Eclypsium and NVIDIA are working on infrastructure verification and continuous monitoring through the Open Agent Safety Platform. Separately, Cloudflare reported that more than 50 families of AI penetration-testing frameworks scanned its customers in the three days before its report, using thousands of IP addresses. ARTEX and its forks made up more than 60% of known orchestrators as of November 12; traffic from these tools rose more than tenfold after ARTEX’s creator took the project down on October 8, and more than 30 times since October 1. Cloudflare recommends defenses including a web application firewall, anomaly detection, bot protection, identity-gated access, validated schemas and threat intelligence.
Google’s Allen says Kubernetes can support agentic AI without companies rebuilding their platforms. Open, collaborative projects aim to reclaim computing resources when agents sit idle waiting for people, making workloads more efficient while retaining their benefits. In scientific research, a review proposes “materials world models” that link material changes, experimental evidence and scientific interpretation over time. The authors describe five challenges and five capabilities, and connect physical models, digital twins and AI agents with automated labs and intelligent instruments. They recommend early demonstrations in alloy aging, phase transformations, battery degradation and catalytic surface evolution, while emphasizing that physical testing and human collaboration remain essential.
Adoption is moving faster than training and workplace rules in medicine. A Doximity survey of 3,379 U.S. physicians, published November 12, 2026, found that 64% of radiologists use AI daily, the third-highest rate among 20 specialties. Forty-five percent said AI saves one to five hours of administrative work weekly, but only 17% said it helps a lot with their skills and knowledge. Three in four had found factual errors in AI answers, two in three regularly check AI-generated clinical information, and only 28% reported clear workplace AI policies. At the University of Birmingham, 96.3% of 80 medical and health science students said they use generative AI, but only 40% had formal training. The exploratory, single-institution study also found links between study program and disclosure, and between training and trust. None raised AI’s environmental footprint; policies at five UK universities did not address sustainability, despite separate sustainability commitments at each. The authors call for AI literacy courses and sustainability-aware policies.
A Cooperative AI initiative aims to give co-operatives a role in shaping AI rules, governance, tools and how value is shared. ICA, Legacoop and PICO Foundation hosted an international bootcamp in Rome from October 12 to 16, bringing together representatives from 18 countries across five continents. The programme seeks shared knowledge and tools grounded in co-operative identity, amid concerns that workers and communities may bear AI’s costs while benefits, data and technological power remain concentrated. In other campus news, a Chapel Hill roundup says UNC officials are promoting greater AI use, while Carolina football prepares to play Louisville. UNC men’s and women’s soccer advanced to the ACC finals; field hockey and cross country began NCAA competition, and both basketball teams prepared for non-conference games. The roundup also mentions a Carrboro High School incident, a restaurant seeking donations and a call to remember UNC graduate students.
Key Takeaways
- North Carolina’s 2026 AI roadmap assigns owners and deadlines, but public notice rights and AI-specific challenge and correction processes remain unresolved or advisory.
- Eclypsium, with NVIDIA’s support, demonstrated that attackers with privileged access can alter AI model weights in memory and change live outputs without changing stored files; the companies are working on verification and monitoring through the Open Agent Safety Platform.
- Cloudflare observed more than 50 AI penetration-testing framework families scanning customers in three days. ARTEX and forks accounted for over 60% of known orchestrators as of November 12, with traffic up more than tenfold after October 8 and more than 30 times since October 1.
- Google’s Allen says Kubernetes projects are working to reclaim computing resources while AI agents wait for people, without requiring companies to rebuild their platforms.
- A materials world models review proposes linking material behavior, experiments and scientific interpretation, with early applications including alloy aging, phase transformations, battery degradation and catalytic surfaces.
- Doximity found 64% of surveyed radiologists use AI daily, while 75% of physicians had found factual errors in AI answers and just 28% reported clear workplace AI policies.
- At the University of Birmingham, 96.3% of 80 surveyed medical and health science students use generative AI, but only 40% had formal training; five UK university AI policies reviewed did not address environmental sustainability.
- A Cooperative AI programme, launched with an October 12–16 Rome bootcamp involving 18 countries, aims to help co-operatives shape AI governance, tools and value-sharing.
- UNC officials are promoting campus AI use; the Chapel Hill roundup also reports soccer teams advancing to ACC finals and other Tar Heel teams beginning NCAA or non-conference competition.
A framework measures whether AI policies let people challenge decisions
The Society-Centered AI Governance Framework helps compare AI policies by asking whether people can learn that AI affected a decision, challenge it and get it corrected. Researchers applied it to North Carolina, interviewing nine AI Leadership Council members before and after the state’s 2026 AI roadmap and reviewing 49 governance documents from 19 states. They found that North Carolina has visibility rules, but notice is an agency duty rather than a right people can use, while AI-specific challenges and corrections remain advisory. The roadmap assigns owners and deadlines for these protections, but the gaps remain open.
Eclypsium warns that hardware attacks can alter AI model outputs
Eclypsium, with support from NVIDIA, showed how attackers with privileged system access could change an AI model while it runs, without changing its stored files or triggering traditional integrity checks. In the demonstration, a compromised data processing unit, or DPU, and weak memory isolation let attackers alter model weights in memory and change its answers during live inference. Eclypsium says this reveals an AI infrastructure assurance gap because standard checks may verify saved files but miss changes in hardware and firmware. The company is working with NVIDIA on infrastructure verification and continuous monitoring through the Open Agent Safety Platform.
Materials world models aim to make AI a lasting scientific partner
A review proposes materials world models that link how materials change, experimental evidence and scientific interpretation over time. The authors describe five challenges, including multiscale behavior, changes tied to processing history and incomplete observations, and outline five capabilities such as tracking evidence, adapting to new results and revising interpretations. Their framework connects a “soft” foundation of physical models, digital twins and AI agents with a “hard” foundation of automated laboratories and intelligent instruments. The authors recommend early demonstrations in alloy aging, phase transformations, battery degradation and catalytic surface evolution, while stressing that human collaboration and physical testing remain essential.
Kubernetes projects aim to reclaim resources when AI agents idle
Google’s Allen says Kubernetes can support agentic AI workloads without requiring companies to rebuild their platforms. Agents can sit idle while waiting for people, and new efforts aim to return those unused computing resources during the wait. Allen described the projects as open, flexible and collaborative, with Google supporting community-driven open source work. He said the goal is to make agent workloads more efficient while still enabling their benefits.
Cloudflare urges defenses against rapidly spreading AI hacking tools
Cloudflare says more than 50 families of AI penetration-testing frameworks tried to scan its customers worldwide in the three days before its report, with thousands of IP addresses running them. The tools use AI agents to automate offensive security work, and ARTEX and its forks accounted for more than 60% of known orchestrators as of November 12. Traffic from these tools grew more than tenfold after ARTEX’s creator took down the original project on October 8, and more than 30 times since October 1. Cloudflare recommends a six-part Modern Security Model that includes a web application firewall, anomaly detection, bot defense, identity-gated access, validated schemas and threat intelligence.
Doximity survey finds 64% of radiologists use AI daily
A Doximity survey of 3,379 U.S. physicians found that 64% of radiologists use AI every day, the third highest rate among the 20 specialties surveyed. Only 17% said AI helps a lot with their skills and knowledge, while 45% said it saves them one to five hours of administrative work each week. Three in four physicians said they had found factual errors in AI answers, and two in three said they regularly check AI-generated clinical information. The survey, published November 12, 2026, also found that only 28% of respondents had clear workplace policies for AI use.
Nearly all surveyed medical students use AI despite limited training
A survey at the University of Birmingham found that 96.3% of 80 medical and health science students use generative AI, while only 40% had received formal training. The study also found links between students’ study programs and whether they disclose their AI use, and between training and trust in AI tools. None of the students raised AI’s environmental footprint on their own. A review of policies at five UK universities found that none of their AI policies addressed environmental sustainability, despite separate sustainability commitments at all five. The authors called for AI literacy courses and sustainability-aware policies, while noting that the study was exploratory and limited to one institution.
Cooperative AI initiative aims to put people and communities first
The Cooperative AI programme aims to help co-operatives shape AI rules, governance, tools, and the sharing of value. It seeks to build shared knowledge, connect co-operative experts, and develop tools based on co-operative identity. From October 12 to 16, ICA, Legacoop, and PICO Foundation hosted an international bootcamp in Rome with representatives from 18 countries across five continents. The initiative responds to concerns that communities and workers may bear the costs of AI development without sharing its benefits, while data and technological power remain concentrated. Its goal is to support innovation that puts people, the environment, and communities ahead of technology.
UNC sports and campus AI featured in Chapel Hill news roundup
The news roundup covers Carolina football preparing to play Louisville and UNC officials promoting greater use of AI on campus. It also mentions an incident at Carrboro High School, a Chapel Hill restaurant seeking donations, and a call to remember UNC graduate students. In sports, UNC men’s and women’s soccer advanced to the ACC finals. Tar Heel field hockey and cross country began NCAA tournament competition, while the Carolina men’s and women’s basketball teams prepared for non-conference games.
Sources
- Between Niti and Nyaya: A Society-Centered Framework for AI Governance
- Eclypsium spotlights AI hardware attack blind spot
- Materials World Models Could Turn AI From a Prediction Tool Into a Persistent Scientific Partner
- Agentic AI workloads push Kubernetes to return idle resources
- Adapting to AI penetration testing orchestrators: The Modern Security Model
- Daily AI use high among radiologists, may help hone skills
- Medical Students Widely Use AI Despite Training and Governance Gaps
- Co-operative AI: A co-operative vision for a fairer, more inclusive and democratic Artificial Intelligence
- The Morning News: UNC Football, UNC Soccer, UNC Artificial Intelligence
Comments
Please log in to post a comment.