Palantir Builds Sovereign AI for US Government Agencies with NVIDIA Nemotron

Palantir has announced a new intelligence engine built on NVIDIA's Nemotron open models. Tailored to the needs of US government agencies, it runs models on NVIDIA accelerated computing in air-gapped environments (fully isolated from external networks).

Government agencies can customize Nemotron models on their own infrastructure, train them on their own data, and fully own the resulting models, including the weights. Palantir's sovereign AI OS (built on AIP, Ontology, Foundry, and Apollo) guarantees data authentication and auditability.

With roughly 3 million civilian employees, the US government is effectively one of the largest "companies" in the world. Its wide-ranging responsibilities, from food safety to highway maintenance, mean the potential value of AI-driven efficiency is enormous. Another notable feature is a "data flywheel" structure: as deployment progresses, the models are continuously improved with new data and feedback.

The advantages highlighted for open models include trustworthiness (biases and vulnerabilities can be discovered through independent review), deployability in regulated environments such as the financial industry (customizability), and cost savings.

School Misstrike in US Military's AI Target Selection—Critical Flaws in Targeting Infrastructure Exposed

It has come to light that when the US military used AI to select thousands of targets, an annotation (a note) indicating that one target was a school was overlooked, and a missile strike was carried out. An investigation into a missile attack on a school in Iran revealed serious flaws in the US military's target selection infrastructure.

AI is expected to fill gaps of this kind, yet the current system fails to properly convey even human annotations. How AI should integrate and prioritize existing metadata and annotations when generating target lists has become an urgent issue.

This incident underscores that, even as AI use in the military expands, human oversight and the pathways by which information is transmitted remain a fatal bottleneck.

Cognition Releases New Coding Benchmark "Frontier Code"; Kimi and GLM Results Draw Attention

Cognition, an AI coding company, has released a new coding benchmark called "Frontier Code." It is designed to evaluate practical coding abilities that existing benchmarks have struggled to measure.

Results for various models on this benchmark are already drawing attention in the community, with the performance of China's Kimi (月之暗面) and GLM (智譜AI) emerging as a hot topic. The LocalLLaMA community is actively discussing how both models fare on Frontier Code.

As existing benchmarks like SWE-bench approach saturation, expectations are high that this will serve as a new standard capable of differentiating models on harder, "frontier"-level tasks.

Omen AI Raises $31 Million—Using AI to Monitor Bacterial Contamination in Data Center Cooling

Omen AI has raised $31 million (approximately 4.8 billion yen) in a Series A round. The company works on monitoring chip coolant and preventing bacterial outbreaks inside data centers.

As AI data centers consume more power and generate more heat, liquid cooling is becoming increasingly important—but it is known that bacteria can grow within coolant circulation systems and cause serious damage to infrastructure. Omen AI offers a solution that uses AI to monitor coolant conditions in real time, detecting and preventing bacterial growth early.

NBER Paper: AI and the Collapse of the WWW

The National Bureau of Economic Research (NBER) has published a working paper (w35344) titled "AI and the Collapse of the Www." The paper analyzes the structural impact of AI on the web ecosystem, and discussion has already begun on Hacker News.

It appears to be an academic examination of how the growth of AI-generated content, intensified data extraction by crawlers, and shifting search traffic are transforming the web's traditional information ecosystem. The details require reading the paper itself, but it presents an important discussion of the long-term impact of AI adoption on the web as a public information infrastructure.

---