Category intelligence

AI News Briefing — February 26, 2026

20 current items analyzed and ranked.

Executive synthesis

AI News Summary

Anthropic dominates this cycle's headlines with a high-stakes showdown against the Pentagon: Defense Secretary Pete Hegseth threatened to invoke the Defense Production Act unless Anthropic grants unfettered military access to Claude, including for autonomous lethal operations and domestic surveillance. Separately, Anthropic reportedly downgraded its AI safety policy, abandoning its commitment to only ship models it deems safe—a seismic shift for the industry's safety standard-bearer.

  • Nvidia reported $62.3B in quarterly data center revenue (75% YoY growth), reinforcing that AI infrastructure investment continues to accelerate despite bubble fears
  • Liquid AI released LFM2-24B-A2B, a novel hybrid architecture blending attention and convolution layers to dramatically reduce memory overhead—a meaningful efficiency breakthrough
  • Google Gemini launched agentic capabilities on the Samsung Galaxy S26, autonomously booking Uber rides and ordering DoorDash meals—among the first mainstream deployments of consumer agentic AI
  • A federal judge dismissed xAI's lawsuit against OpenAI, finding no evidence of misconduct
  • CuspAI (founded by Max Welling, advised by Hinton and LeCun) raised $100M for AI-driven materials discovery
  • Meta open-sourced GCM for GPU cluster monitoring, while its AI moderation was criticized for flooding investigators with junk reports

Key Themes

AI Safety & Military Use · 3AI Infrastructure & Economics · 3Agentic AI · 4Model Architecture & Efficiency · 1AI Policy, Law & Ethics · 5AI for Science · 1

Primary evidence

Top Ranked Signals

News Ars Technica - All content Feb 25

Pete Hegseth tells Anthropic to fall in line with DoD desires, or else

By George Hammond and Steff Chávez, Financial Times

95 score
AI Analysis

Building on Reddit coverage from two days ago about the initial meeting, US Defense Secretary Pete Hegseth has given Anthropic until Friday to grant the military unfettered access to Claude for all lawful applications—including domestic surveillance and lethal autonomous operations—or face being cut from the DoD supply chain. Hegseth also threatened to invoke the Defense Production Act, a Cold War-era compulsory measure. This represents an unprecedented confrontation between AI safety commitments and national security demands.

US Defense Secretary Pete Hegseth has threatened to cut Anthropic from his department’s supply chain unless it agrees to sign off on its technology being used in all lawful military applications by Friday. The threat is the latest escalation in a feud between Anthropic and the department, triggered by the AI group’s refusal to give unfettered access to its models for classified military use, including domestic surveillance and deadly missions with no direct human control. Hegseth summoned Anthro
AI SafetyMilitary AIGovernment PolicyAnthropic
News AI (artificial intelligence) | The Guardian Feb 25

US military leaders pressure Anthropic to bend Claude safeguards

By Nick Robins-Early

93 score
AI Analysis

Following yesterday's Reddit discussion of the Pentagon's dual approach to xAI and Anthropic, Guardian's coverage of the same Anthropic-Pentagon confrontation adds context that Hegseth gave Amodei until end of day Friday to comply or face penalties. The dispute centers on Anthropic's refusal to remove safeguards from Claude for classified military use. Anthropic, which markets itself as the most safety-forward AI lab, faces an existential policy dilemma.

Anthropic presents itself as most safety-forward AI firm and Pentagon has threatened penalties if it does not yieldUS military leaders including Pete Hegseth, the defense secretary, met with executives from the artificial intelligence firm Anthropic on Tuesday to hash out a dispute over what the government will be able to do with the company’s powerful AI model. Hegseth gave Dario Amodei, the Anthropic CEO, until the end of the day on Friday to agree to the department’s terms or face penalties,
AI SafetyMilitary AIGovernment PolicyAnthropic
90 score
AI Analysis

As first announced on Social yesterday by Anthropic, Anthropic has reportedly downgraded its AI safety policy, retreating from its prior commitment to only release models it classified as safe. The change comes amid intense market competition and government pressure. This signals a potential weakening of safety norms across the frontier AI industry.

The AI vendor previously committed to releasing only models it classified as safe.
AI SafetyAnthropicIndustry Norms
News AI (artificial intelligence) | The Guardian Feb 25

Nvidia quarterly earnings show immunity to AI bubble fears as it cashes in on data center boom

By Nick Robins-Early

85 score
AI Analysis

Nvidia reported $62.3B in data center revenue for the quarter, representing 75% year-over-year growth and once again surpassing Wall Street expectations. The company remains the world's most valuable publicly traded company, with its GPUs serving as the backbone of the global AI infrastructure buildout.

Chipmaker’s quarterly earnings surpassed Wall Street’s expectations every quarter for multiple years nowNvidia released its quarterly earnings on Wednesday, with the chipmaker revealing higher than expected revenues and extending its yearslong streak of surpassing Wall Street’s sky-high expectations.The company receives the vast majority of its revenue from its data center business, which has been buoyed by the tech industry’s immense investment into AI infrastructure. On Wednesday, Nvidia repor
AI InfrastructureNvidiaFinancial MarketsData Centers
78 score
AI Analysis

Liquid AI released LFM2-24B-A2B, a 24B parameter model using a novel hybrid architecture that combines attention layers with convolution-based 'base' layers at a 1:3 ratio. This approach dramatically reduces KV cache memory requirements while maintaining strong performance, targeting the efficiency bottlenecks that plague standard Transformer architectures at scale.

The generative AI race has long been a game of ‘bigger is better.’ But as the industry hits the limits of power consumption and memory bottlenecks, the conversation is shifting from raw parameter counts to architectural efficiency. Liquid AI team is leading this charge with the release of LFM2-24B-A2B, a 24-billion parameter model that redefines what we should expect from edge-capable AI. www.liquid.ai/blog/lfm2-24b-a2b The ‘A2B’ Architecture: A 1:3 Ratio fo
Model ArchitectureEfficiencyNew Model ReleaseResearch
News Feed: Artificial Intelligence Latest Feb 25

Gemini Can Now Book You an Uber or Order a DoorDash Meal on Your Phone. Here’s How It Works

By Julian Chokkattu

74 score
AI Analysis

Google's Gemini can now autonomously perform tasks within third-party mobile apps, including booking Uber rides and ordering DoorDash meals, launching first on the Samsung Galaxy S26. This represents a concrete step toward agentic AI operating in real-world consumer workflows.

Starting with the Samsung Galaxy S26, Google’s Gemini can automate tasks in popular mobile apps. We got a live demo of the new feature in action.
Agentic AIGoogle GeminiConsumer ProductsSamsung
News Ars Technica - All content Feb 25

Musk has no proof OpenAI stole xAI trade secrets, judge rules, tossing lawsuit

By Ashley Belanger

68 score
AI Analysis

A federal judge dismissed xAI's trade secret lawsuit against OpenAI, ruling that Musk's company provided no evidence that OpenAI induced former xAI employees to steal secrets or that any stolen secrets were used. The ruling undermines Musk's legal offensive against OpenAI.

Elon Musk appears to be grasping at straws in a lawsuit accusing OpenAI of poaching eight xAI employees in an allegedly unlawful bid to access xAI trade secrets connected to its data centers and chatbot, Grok. In a Tuesday order granting OpenAI's motion to dismiss, US District Judge Rita F. Lin said that xAI failed to provide evidence of any misconduct from OpenAI. Instead, xAI seemed fixated on a range of alleged conduct of former employees. But in assessing xAI's claims, Lin said that xAI fail
LegalOpenAIxAIElon Musk
67 score
AI Analysis

CuspAI, co-founded by ML pioneer Max Welling, is applying equivariant neural networks and diffusion models to search the space of all possible materials for climate-relevant applications. The company raised a $100M Series A and is rumored to have reached unicorn valuation, with advisors including Hinton and LeCun.

Editor’s note: CuspAI raised a $100m Series A in September and is rumored to have reached a unicorn valuation. They have all-star advisors from Geoff Hinton to Yann Lecun and team of deep domain experts to tackle this next frontier in AI applications.In this episode, Max Welling traces the thread connecting quantum gravity, equivariant neural networks, diffusion models, and climate-focused materials discovery (yes, there is one!!!).We begin with a provocative framing: experiments as comput
AI for ScienceFundingMaterials DiscoveryDiffusion Models
News AI (artificial intelligence) | The Guardian Feb 25

Meta’s AI sending ‘junk’ tips to DoJ, US child abuse investigators say

By Katie McQue

64 score
AI Analysis

Meta's AI-powered content moderation is generating large volumes of useless child sexual abuse reports to the DOJ, draining investigator resources and slowing real cases. Testimony from ICAC taskforce agents in New Mexico's trial against Meta revealed the reports are largely 'junk.'

Officers say flood of low-quality reports is draining resources and slowing cases amid New Mexico lawsuitMeta’s use of artificial intelligence software to moderate its social media platforms is generating large volumes of useless reports about cases of child sexual abuse, which are draining resources and hindering investigations, said officers from the US Internet Crimes Against Children (ICAC) taskforce.“We get a lot of tips from Meta that are just kind of junk,” Benjamin Zwiebel, a special age
AI ModerationMetaChild SafetyAI Failures
62 score
AI Analysis

Meta open-sourced GCM (GPU Cluster Monitoring), a specialized toolkit for monitoring hardware health in large-scale AI training clusters. The tool addresses 'silent' hardware failures that can corrupt training runs on infrastructure with thousands of GPUs.

While the tech folks obsesses over the latest Llama checkpoints, a much grittier battle is being fought in the basements of data centers. As AI models scale to trillions of parameters, the clusters required to train them have become some of the most complex—and fragile—machines on the planet. Meta AI Research team just released GCM (GPU Cluster Monitoring), a specialized toolkit designed to solve the ‘silent killer’ of AI progress: hardware instability at scale. GCM is a blueprint
Open SourceAI InfrastructureMetaGPU Clusters
News AI (artificial intelligence) | The Guardian Feb 25

Facial recognition error prompts police to arrest Asian man for burglary 100 miles away

By Robert Booth and Mark Wilding

58 score
AI Analysis

UK police wrongfully arrested a man based on a facial recognition error that confused him with a different person of South Asian heritage located 100 miles away. He was held for nearly 10 hours before release. The case highlights persistent racial bias in facial recognition systems deployed by law enforcement.

Exclusive: Alvi Choudhury claiming damages against Thames Valley police after biased technology confused him with man looking ‘10 years younger’Police arrested a man for a burglary in a city he had never visited after face scanning software deployed across the UK confused him with another person of south Asian heritage.Alvi Choudhury, 26, a software engineer, was working at the home he shares with his parents in Southampton in January when police knocked on his door, handcuffed him and held him
AI BiasFacial RecognitionCivil LibertiesLaw Enforcement
News aibusiness Feb 25

Google Labs adds Agentic AI Capabilities to Opal

By Scarlett Evans

57 score
AI Analysis

Google Labs added agentic AI capabilities to Opal, enabling goal-driven task planning and autonomous execution. Details are limited but this expands Google's agentic AI product portfolio.

The interactive agent enables goal-driven task planning and execution.
Agentic AIGoogleProduct Launch