AMD and Oak Ridge Open Lux Supercomputer for Science

Across healthcare research, reviews point to AI’s potential alongside a persistent gap between strong test results and dependable clinical use. In a two-center study of 242 patients undergoing capsule endoscopy, AI reduced image-review time from 38.1 minutes to 18.26 minutes, with 95.3% sensitivity and 96.5% specificity. But a review of AI for lung cancer biomarkers found 67% of 82 studies at high overall risk of bias. Retinal AI studies also reported strong results—including 93% sensitivity and 94% specificity for the i-ROP system on a separate 100-image test—yet differences between studies and clinical workflow challenges limit what those results establish. Reviews of liver cancer AI describe promising biopsy scoring, cancer classification and tissue analysis, but say evidence of improved patient outcomes remains limited.

Other medical and scientific reviews stress validation and oversight. AI methods can help researchers discover and characterize nanomaterials, but data limitations, reproducibility and regulatory barriers complicate translation beyond the lab. Biological digital twins could model living systems and potentially reduce animal testing; toxicologist Thomas Hartung says they need to represent a specific system, update with data, follow biological rules and make experimentally testable predictions. He calls for shared standards, independent testing and oversight, particularly for self-updating systems. In orthopedics, researchers likewise urge external testing, probability calibration and attention to clinical workflow rather than reliance on retrospective results alone.

AI infrastructure and research collaborations are expanding. AMD, Oak Ridge National Laboratory and the U.S. Department of Energy have opened Lux to scientific users. Ranked 14th on TOP500, the supercomputer combines AMD EPYC CPUs, Instinct GPUs and Pensando networking and will support DOE’s Genesis Mission, including work on fusion power plant design, materials, manufacturing and Earth system predictions. Separately, the Cornell High Energy Synchrotron Source collaboration won a 2026 HPCwire Editors’ Choice Award for connecting AI, high-performance computing and scientific instruments to enable autonomous experiments. Indiana University has opened a Bloomington center for research in cybersecurity, electromagnetic spectrum operations, AI, microelectronics and biotechnology, linking university researchers with government and industry.

Several developments focus on tools and deployment. Microsoft says its Perception security-agent platform is generally available, with four blue-team agents for alert triage, attack investigations, detection writing and threat intelligence. The company says its Alert Triage Agent can handle about 80% of alerts in a typical Defender queue, based on roughly 8 million alerts. A web-app penetration-testing agent remains in preview, while green-team playbooks span Microsoft, AWS and Google Cloud. Microsoft is also previewing detection for 60 types of local AI agents, including Claude Code, OpenAI Codex and GitHub Copilot. A separate comparison examines 13 agent gateways that manage tool access through MCP, agent-to-agent communication through A2A and connections to business systems; products vary in deployment, maturity and price transparency.

Research on language models and large-scale data processing offers other approaches. BOS Summary changes how a model uses its beginning-of-sequence token: during one transformer layer, that token can attend to all visual and text input, providing context through decoding without extra parameters, modules or decoding-time overhead. Tests across model families and eight text-only and multimodal benchmarks found gains over direct and Chain-of-Thought prompting and other attention-sink methods. Astronomers, meanwhile, plan a virtual observatory combining data from the 10-year Rubin Observatory survey, Europe’s Euclid telescope and the upcoming Roman telescope. AI and cloud computing are expected to help process and combine the data into sharper views for studies of dark matter, dark energy, supernovae, asteroids and exoplanets.

Commercial and public uses raise questions about trust and responsibility. Liquid Studios, formerly Inception AI, has shifted away from producing its own podcasts—once more than 15,000—and now helps businesses turn text into audio and audio into podcast videos. CEO Jeanine Wright cited concerns about AI advertising, content quality and maintaining a back catalog of more than 350,000 episodes; the company also removed health and wellness episodes after AI hosts presented themselves as medical or mental-health professionals. It raised a $5 million seed round led by Activate Venture Partners in January, grew from eight employees to 11, and lost co-founders Katie Brown and William Corbin. In political advertising, marketing professor Kristen Walker says AI makes fabricated messages harder to distinguish from real ones. AI ads became more common in the 2026 midterms after emerging in 2024; 31 states have laws prohibiting deepfakes or requiring disclosures, while questions about acceptable uses remain.

Two announcements offer little or no detail: OPI’s November 19, 2026, listing says ECI introduced an AI solution for small businesses, but the full article is paywalled and does not explain the product. The supplied articles also do not report specific new AI product or investment announcements from Google, Meta, Amazon or OpenAI; Google and OpenAI appear in coverage of Microsoft’s cloud playbooks and agent detection, respectively.

Key Takeaways

  • In a 242-patient, two-center capsule-endoscopy study, AI cut review time from 38.1 to 18.26 minutes, with 95.3% sensitivity and 96.5% specificity.
  • A review of 82 AI studies on non-small cell lung cancer biomarkers found 67% at high overall risk of bias; medical reviews broadly call for stronger validation and real-world outcome evidence.
  • AMD, Oak Ridge National Laboratory and the U.S. Department of Energy opened Lux, a TOP500 No. 14 supercomputer, for scientific users and DOE’s Genesis Mission.
  • Microsoft’s generally available Perception platform includes four security agents; Microsoft says its Alert Triage Agent can handle about 80% of typical Defender-queue alerts, based on roughly 8 million alerts.
  • Microsoft’s previewed security tools detect 60 types of local AI agents, including OpenAI Codex and GitHub Copilot; its green-team playbooks span Microsoft, AWS and Google Cloud.
  • Liquid Studios shifted from making its own podcasts to business audio and video tools, after raising a $5 million seed round led by Activate Venture Partners.
  • AI-generated political ads and deepfakes raise trust concerns; 31 states have laws prohibiting deepfakes or requiring disclosures.
  • A planned virtual observatory will combine Rubin, Euclid and the upcoming Roman telescope surveys, using AI and cloud computing to process and combine their data.
  • Reviews of biological digital twins, nanotechnology and orthopedic AI emphasize independent testing, reproducibility, standards, oversight and fit with real-world use.
  • OPI’s report on ECI’s AI solution for small businesses provides no product details because the article is available only to paying members.

AI and Confocal Imaging Aim to Improve Small Bowel Diagnosis

A review in Frontiers in Medicine examines how artificial intelligence and confocal laser endomicroscopy could help diagnose hard-to-find small intestine disorders. Capsule endoscopy can capture tens of thousands of images, and doctors may need 30 minutes to two hours to review them. In a two-center study of 242 patients, AI cut review time from 38.1 minutes to 18.26 minutes while reaching 95.3 percent sensitivity and 96.5 percent specificity. The review also notes that false positives, limited testing on rare conditions, and a lack of explainable AI remain important concerns.

AI Speeds Nanotechnology Research but Needs Stronger Testing

A review in Health Nanotechnology examines how artificial intelligence can help discover, make, and study nanomaterials, which measure between 1 and 100 nanometers. Machine learning can find patterns between a material’s ingredients and its properties, helping researchers improve experiments and reduce some computational work. The review describes tools including neural networks, random forests, support vector machines, Gaussian process regression, and graph neural networks. It also identifies data limits, reproducibility problems, and regulatory barriers that make it difficult to move laboratory findings into clinical and industrial use.

Review Finds High Bias Risk in AI Studies of Lung Cancer Biomarkers

A systematic review in Discover Artificial Intelligence examined 82 studies on machine learning and deep learning for non-small cell lung cancer biomarkers. The researchers found that 67 percent of the models had a high overall risk of bias, raising doubts about whether their reported performance can be trusted in patient care. The team searched seven databases and covered studies published through May 31, 2026. Convolutional neural networks were the most common approach, appearing in 28 studies, or 34 percent of the total.

Biological Digital Twins Show Promise but Need Stronger Validation

A theoretical analysis in Frontiers in Artificial Intelligence explores how AI could build biological digital twins that model living systems and may help reduce animal testing. Toxicologist Thomas Hartung says a true twin must represent a specific biological system, update with new data, follow biological rules, and make predictions that experiments can test. The analysis describes models at different scales, including cell systems using tools such as Geneformer, scGPT, and scFoundation. Hartung warns that agreed standards, independent testing, and oversight of systems that update themselves are needed for regulatory acceptance.

Retinal AI Shows Strong Accuracy but Faces Real-World Hurdles

A review in Frontiers in Artificial Intelligence assessed 18 studies of AI for retinopathy of prematurity, diabetic retinopathy, and retinal vein occlusion. The researchers, led by Fapianey J. Alexandre at Meharry Medical College, reviewed literature found in PubMed and Google Scholar through August 2026. The studies tested several types of retinal images, and some models showed high accuracy, including the i-ROP system, which reached 93 percent sensitivity and 94 percent specificity on a separate 100-image test. Differences in study methods and limits in the literature search prevented the authors from combining results in a meta-analysis, and high accuracy alone does not ensure that tools will fit into clinical care.

AI Shows Promise in Liver Cancer Diagnosis but Lacks Outcome Evidence

A review found that artificial intelligence models can support liver cancer screening, tissue analysis, and treatment planning, but strong evidence that they improve real-world patient outcomes is still lacking. Researchers see growing need for better tools as liver cancer linked to metabolic dysfunction and alcohol use rises, including cases in people without cirrhosis. AI systems have shown strong results in tasks such as scoring liver biopsies, distinguishing hepatocellular carcinoma from cholangiocarcinoma, and identifying tissue features and mutations. The review also describes a shift toward foundation models trained on large collections of tissue slides, while noting that technical accuracy alone does not prove better care.

Orthopedic AI Research Turns to Real World Testing and Workflow Fit

A review in Knee Surgery & Related Research says orthopedic AI research has grown quickly, but routine clinical use remains limited. The authors point to concerns including models that are hard to understand, weaker performance in new health care settings, and poor fit with everyday clinical workflows. Orthopedic tools use electronic medical records, medical images, and clinical text, with methods ranging from traditional machine learning to deep learning and large language models. The authors say the field should focus more on independent external testing, probability calibration, and implementation, rather than only reporting results from retrospective studies.

AMD and Oak Ridge Open Lux Supercomputer for AI Science

Lux, a supercomputer developed by AMD, Oak Ridge National Laboratory, and the U.S. Department of Energy, is now open to scientific users. Ranked 14th on the TOP500 list, it supports the DOE’s Genesis Mission with an AI Factory for scientific research and engineering. Lux combines AMD EPYC CPUs, Instinct GPUs, and Pensando networking. Researchers will use it for projects involving fusion power plant design, materials development, manufacturing, and Earth system predictions.

CHESS Collaboration Wins Award for AI Driven Experiments

The Cornell High Energy Synchrotron Source collaboration has won the 2026 HPCwire Editors’ Choice Award. The award recognizes its work connecting artificial intelligence, high-performance computing, and scientific instruments across the country to enable autonomous experiments. The collaboration brings these tools together to support scientific discovery and innovation. The recognition highlights the role of cross-discipline teamwork in advancing AI-driven research.

Indiana University Opens National Security Research Center

Indiana University opened a center in Bloomington on Friday, Nov. 13, focused on national security research. Its work will cover cybersecurity, electromagnetic spectrum operations, artificial intelligence, microelectronics, and biotechnology, connecting university researchers with government and industry partners. Scott Bullock, director of the Technology, said the center will serve as an entry point for defense research and innovation. The university also expects it to attract federal research investment to Indiana and help prepare students for careers in national security and advanced technology.

Liquid Studios shifts from AI podcasts to business content tools

Liquid Studios, formerly known as Inception AI, has stopped producing and earning money from its own podcasts, which once numbered more than 15,000. The company now helps businesses turn text into audio and make podcast videos from audio, after shifting direction in mid-July. CEO Jeanine Wright said concerns about AI advertising, content quality and the work needed to maintain a back catalog of more than 350,000 episodes helped drive the change. The company also removed health and wellness episodes after concerns about AI hosts presenting themselves as doctors or mental health professionals. Liquid Studios raised a $5 million seed round led by Activate Venture Partners in January and has grown from eight employees to 11, though co-founders Katie Brown and William Corbin have left.

Microsoft makes its Perception AI security agents generally available

Microsoft announced at its Ignite 2026 conference that Perception is now generally available for coordinating AI agents on security tasks. Its four blue team agents handle alert triage, attack investigations, detection writing and threat intelligence. Microsoft said the Alert Triage Agent can handle about 80% of alerts in a typical Defender queue, based on an analysis of roughly 8 million alerts. The Recon Agent is also generally available, while a web app penetration testing agent is in preview; new green team playbooks work across Microsoft, AWS and Google Cloud. Microsoft is also previewing tools that detect 60 types of local AI agents, including Claude Code, OpenAI Codex and GitHub Copilot.

AI political ads raise new concerns about trust and authenticity

Kristen Walker, a marketing professor at California State University, Northridge, said AI is making it harder to tell real political messages from fabricated ones. Campaigns use AI to quickly and cheaply make flyers, infographics, synthetic videos and deepfakes, which may help candidates without large budgets compete. Walker said the technology also raises concerns about truth and public trust, as AI-generated ads became more common during the 2026 midterms after beginning to emerge in the 2024 election cycle. According to the article, 31 states have laws that prohibit deepfakes or require disclosures. Walker expects more rules in the future, but said people still need to decide what uses of AI in political advertising are acceptable.

BOS Summary gives language models more useful context

BOS Summary is a training-free method that changes how language and vision-language models use the beginning-of-sequence, or BOS, token. During one transformer layer, it lets the BOS token attend to all visual and text input, while other tokens keep standard causal attention. This gives the model a context summary it can use throughout decoding, without adding parameters, extra modules or decoding-time overhead. Tests across multiple model families and eight text-only and multimodal benchmarks found improvements over direct and Chain-of-Thought prompting, as well as recent methods that change attention sinks. The researchers also found that making BOS specific to the input reduces its dominance and shifts attention toward meaningful prompt content.

ECI announces an AI solution for small businesses

OPI published an article titled ā€œECI introduces AI solution for SMBsā€ on November 19, 2026. The page identifies the story as being about North America, but the article text is available only to paying members. It does not provide details about the AI solution or how it works. OPI describes itself as a source of news, analysis and resources for the global business supplies industry.

Astronomers plan an AI-powered virtual supertelescope

Astronomers want to combine data from three large telescope surveys into a virtual observatory. The surveys include the 10-year Rubin Observatory project in Chile, the European Euclid telescope in space, and the upcoming Roman telescope. Researchers say AI and cloud computing will help process the huge volume of information and combine telescope images into sharper views. The project could improve studies of dark matter, dark energy, short-lived events such as supernovae, asteroids, and planets beyond our solar system.

13 agent gateways compared for MCP, A2A, and AI traffic

Agent gateways control how AI agents access tools through MCP, communicate with other agents through A2A, and connect to business systems. This comparison ranks 13 gateways using traffic coverage, identity and authorization, deployment choices, maturity, and price transparency. The list includes Kong, Bifrost, Gravitee, AgentCore Gateway, Arcade, and other products, with different strengths for cloud, self-hosted, and enterprise use. It also notes limits such as features that require paid plans, preview support, or missing public pricing.

Sources

NOTE:

This news brief was generated using AI technology (including, but not limited to, Google Gemini API, Llama, Grok, and Mistral) from aggregated news articles, with minimal to no human editing/review. It is provided for informational purposes only and may contain inaccuracies or biases. This is not financial, investment, or professional advice. If you have any questions or concerns, please verify all information with the linked original articles in the Sources section below.

Healthcare AI Clinical validation Medical imaging Capsule endoscopy Lung cancer biomarkers Retinal AI Liver cancer AI bias Biological digital twins Nanomaterials Orthopedic AI High-performance computing AMD Oak Ridge National Laboratory Microsoft security agents AI agent gateways Large language models Astronomy AI-generated content Deepfakes

Comments

Loading...