Wiredframe Radar logo

Category intelligence

AI News Briefing — February 24, 2026

21 current items analyzed and ranked.

Executive synthesis

AI News Summary

OpenAI's Frontier Evals team officially retired SWE-Bench Verified due to score saturation, endorsing SWE-Bench Pro as successor—a milestone signaling coding AI maturity. In a major geopolitical escalation, Anthropic accused DeepSeek, Moonshot AI, and MiniMax of industrial-scale distillation from Claude, following similar charges by OpenAI last month.

Key Themes

AI Benchmarks & Evaluation · 1US-China AI Competition & IP · 1AI Copyright & Legal · 1AI Infrastructure & Energy · 4AI Hardware Innovation · 1Agentic AI & Commerce · 4AI in Creative Industries · 1

Primary evidence

Top Ranked Signals

News Latent.Space Feb 23

⚡️The End of SWE-Bench Verified — Mia Glaese & Olivia Watkins, OpenAI Frontier Evals & Human Data

By Unknown

88 score
AI Analysis

OpenAI's Frontier Evals team has officially discontinued SWE-Bench Verified, the leading coding benchmark, due to score saturation. They are endorsing SWE-Bench Pro as its successor. This signals that frontier models have effectively maxed out the benchmark.

First speakers for AIE Europe and AIEi Miami have been announced. See you there!We’ve been somewhat making tongue in cheek references to the very very minor bumps on SWE-Bench Verified scores every time a new frontier model is released (Opus 4.5 → 4.6 was literally a 0.1% down step), but it is a whole other matter for the original authors of SWE-Bench Verified to make the call to discontinue reporting it. We were excited to have Mia Glaese, original coauthor of SWE-Bench Verified and
benchmarksfrontier evalscoding AIOpenAI
News AI (artificial intelligence) | The Guardian Feb 23

US AI giant accuses Chinese rivals of mass data theft

By Agence France-Presse

87 score
AI Analysis

Anthropic accuses three Chinese AI firms—DeepSeek, Moonshot AI, and MiniMax—of industrial-scale distillation from its Claude chatbot to boost their own models. OpenAI made similar accusations last month. This escalates US-China AI tensions significantly.

Anthropic says three Chinese firms used ‘distillation’ technique to extract information from its Claude chatbotUS artificial intelligence company Anthropic said on Monday it had uncovered campaigns by three Chinese AI firms to illicitly extract capabilities from its Claude chatbot, in what it described as industrial-scale intellectual property theft. OpenAI leveled similar charges last month.Anthropic said DeepSeek, Moonshot AI and MiniMax used a technique known as “distillation” – using outputs
US-China AI competitionintellectual propertyAnthropicDeepSeekdistillation
News Ars Technica - All content Feb 23

AIs can generate near-verbatim copies of novels from training data

By Melissa Heikkilä, Financial Times

85 score
AI Analysis

Studies show top AI models from OpenAI, Google, Meta, Anthropic, and xAI can generate near-verbatim copies of bestselling novels, demonstrating far more memorization than previously claimed. This undermines the industry's core legal defense in copyright lawsuits.

The world’s top AI models can be prompted to generate near-verbatim copies of bestselling novels, raising fresh questions about the industry’s claim that its systems do not store copyrighted works. A series of recent studies has shown that large language models from OpenAI, Google, Meta, Anthropic, and xAI memorize far more of their training data than previously thought. AI and legal experts told the FT this “memorization” ability could have serious ramifications on AI groups’ battle against doz
copyrightAI policymemorizationlegal
News MarkTechPost Feb 23

Taalas is replacing programmable GPUs with hardwired AI chips to achieve 17,000 tokens per second for ubiquitous inference

By Asif Razzaq

75 score
AI Analysis

As first reported in Social yesterday, Toronto startup Taalas is building hardwired AI inference chips that bypass GPU programmability, claiming 17,000 tokens per second by casting model weights directly into silicon. The approach aims to overcome the 'memory wall' bottleneck that limits GPU-based inference.

In the high-stakes world of AI infrastructure, the industry has operated under a singular assumption: flexibility is king. We build general-purpose GPUs because AI models change every week, and we need programmable silicon that can adapt to the next research breakthrough. But Taalas, the Toronto-based startup thinks that flexibility is exactly what’s holding AI back. According to Taalas team, if we want AI to be as common and cheap as plastic, we have to stop ‘simulating’ intellig
AI hardwareinference optimizationstartupschips
News AI (artificial intelligence) | The Guardian Feb 23

New datacentres risk doubling Great Britain’s electricity use, regulator says

By Dan Milmo and Jillian Ambrose

74 score
AI Analysis

UK regulator Ofgem warns that roughly 140 proposed data center projects would require 50GW of electricity—exceeding Britain's entire current peak demand of 45GW. The surge is driven primarily by AI workloads.

Ofgem says about 140 proposed projects, driven by AI use, could require more power than current peak demandThe amount of power being sought by new datacentre projects in Great Britain would exceed the national current peak electricity consumption, according to an industry watchdog.Ofgem said about 140 proposed datacentre schemes, driven by use of artificial intelligence, could require 50 gigawatts of electricity – 5GW more than the country’s current peak demand. Continue reading...
energydata centersAI infrastructureregulationUK
News AI News Feb 23

Mastercard’s AI payment demo points to agent-led commerce

By Muhammad Zulhusni

68 score
AI Analysis

Mastercard demonstrated its first fully authenticated 'agentic commerce' transaction at the India AI Impact Summit 2026, where an AI agent autonomously searched for, selected, and purchased a product. The demo used stored payment credentials within a secure verification framework.

A recent demonstration from Mastercard suggests that payment systems may be heading toward a future where software agents, not people, complete purchases. During the India AI Impact Summit 2026, Mastercard showed what it described as its first fully authenticated “agentic commerce” transaction. In the demo, as reported by Times of India, an AI agent searched for a product, assessed the website, and completed the purchase using stored payment credentials, without the user opening an a
agentic AIpaymentscommerceMastercard
News MarkTechPost Feb 23

VectifyAI Launches Mafin 2.5 and PageIndex: Achieving 98.7% Financial RAG Accuracy with a New Open-Source Vectorless Tree Indexing.

By Michal Sutter

65 score
AI Analysis

VectifyAI launched Mafin 2.5 and open-source PageIndex, achieving 98.7% accuracy on financial RAG tasks using a novel 'vectorless' tree indexing approach. The system preserves document structure critical for financial documents like 10-K filings.

Building a Retrieval-Augmented Generation (RAG) pipeline is easy; building one that doesn’t hallucinate during a 10-K audit is nearly impossible. For devs in the financial sector, the ‘standard’ vector-based RAG approach—chunking text and hoping for the best—often results in a ‘text soup’ that loses the vital structural context of tables and balance sheets. VectifyAI is attempting to close this gap with the launch of Mafin 2.5, a multimodal financial agent, and P
RAGfinance AIopen sourcedocument understanding
News MarkTechPost Feb 23

Beyond Simple API Requests: How OpenAI’s WebSocket Mode Changes the Game for Low Latency Voice Powered AI Experiences

By Michal Sutter

63 score
AI Analysis

OpenAI's Realtime API WebSocket mode collapses the traditional STT→LLM→TTS voice pipeline into a single persistent connection with GPT-4o's native multimodal capabilities. This eliminates inter-hop latency for voice AI applications.

In the world of Generative AI, latency is the ultimate killer of immersion. Until recently, building a voice-enabled AI agent felt like assembling a Rube Goldberg machine: you’d pipe audio to a Speech-to-Text (STT) model, send the transcript to a Large Language Model (LLM), and finally shuttle text to a Text-to-Speech (TTS) engine. Each hop added hundreds of milliseconds of lag. OpenAI has collapsed this stack with the Realtime API. By offering a dedicated WebSocket mode, the platform provide
voice AIOpenAIdeveloper toolsreal-time AI
News AI News Feb 23

Hitachi bets on industrial expertise to win the physical AI race

By Dashveenjit Kaur

58 score
AI Analysis

Hitachi is positioning its industrial manufacturing expertise as a competitive advantage in physical AI, arguing that controlling robots in real-world settings requires deep domain knowledge that pure AI labs lack. The strategy is moving from concept to factory floor deployment.

Physical AI – the branch of artificial intelligence that controls robots and industrial machinery in the real world – has a hierarchy problem. At the top, OpenAI and Google are scaling multimodal foundation models. In the middle, Nvidia is building the platforms and tools for physical AI development. And then there is a third camp: industrial manufacturers like Hitachi and Germany’s Siemens, that are making the quieter but arguably more grounded argument that you cannot train machines to n
physical AIroboticsmanufacturingHitachi
News Ars Technica - All content Feb 23

New Microsoft gaming chief has "no tolerance for bad AI"

By Kyle Orland

57 score
AI Analysis

New Microsoft gaming chief Asha Sharma, promoted after Phil Spencer's departure, pledged 'no tolerance for bad AI' and promised games will remain 'art crafted by humans.' She comes from Microsoft's CoreAI Product group but is distancing from aggressive AI integration.

Last week's surprise departure of Phil Spencer from Microsoft led to the promotion of Asha Sharma, who comes to head Microsoft's gaming division after two years as president of the company's CoreAI Product group. Despite that recent history, Sharma says in a new interview that she has "no tolerance for bad AI" in game development. Speaking with Variety, Sharma noted that "AI has long been part of gaming and will continue to be," before adding that "great stories are created by humans." The inter
gamingMicrosoftAI in creative industriesleadership
News Ars Technica - All content Feb 23

Data center builders thought farmers would willingly sell land, learn otherwise

By Ashley Belanger

55 score
AI Analysis

American farmers across the country are refusing multi-million-dollar offers from tech companies seeking rural land for data centers. Some offers significantly exceed property values, but farmers are unwilling to sell land they've nurtured for decades.

It seems that tech giants eyeing rural zones for data center development have underestimated how attached American farmers have grown to their lands in the decades they've been nurturing them. Across the country, several farmers have firmly rejected eye-popping offers—sometimes in the tens of millions. These offers dwarf the value of their properties, but farmers have refused to put a price on the lands that they love most. In a report on Monday, The Guardian highlighted a handful of cases natio
data centersAI infrastructureland userural impact
News AI (artificial intelligence) | The Guardian Feb 23

Sam Altman defends AI’s energy toll by saying it also takes a lot to ‘train a human’

By Eric Berger

52 score
AI Analysis

Sam Altman defended AI's energy consumption at the India AI Impact Summit by comparing it to the energy needed to 'train a human' over 20 years. He also downplayed concerns about data center water usage.

OpenAI CEO also downplayed concerns about how much water datacenters require at AI summit in IndiaThe OpenAI boss, Sam Altman, has tried to ease concerns about how much power is used by artificial intelligence models by comparing it to the amount of energy required by human development.“People talk about how much energy it takes to train an AI model – but it also takes a lot of energy to train a human,” Altman told the Indian Express recently while in India for the AI Impact summit. “It takes ab
energyOpenAISam Altmansustainability