Short News Archive

Mar 3, 2026

Several US federal agencies are already dropping Anthropic's AI products and switching to competitors like OpenAI. According to Reuters, the shift currently affects the State Department, Treasury Department, Department of Health and Human Services, the Pentagon, and the Department of Housing and Urban Development. President Trump ordered all agencies on Friday to phase out Anthropic products within six months. The Department of Defense had previously classified Anthropic as a supply chain risk and signed a deal with OpenAI.

The switch isn't exactly an upgrade, though - at least not yet. The State Department is replacing Anthropic's Claude models in its internal chatbot with OpenAI's outdated GPT-4.1 model.

Comment Source: Reuters

Matthias Bastian

Mar 3, 2026

Short News

Anthropic entered a $100 million Pentagon competition in early 2026. The company was proposing to use Claude for voice-controlled autonomous drone swarm technology, Bloomberg reports.

The idea was to use Claude to translate a commander's spoken orders into digital instructions and coordinate drone fleets without using AI for autonomous targeting or weapons decisions. Humans would be able to monitor and shut down the system. This approach lines up with Anthropic's position in its ongoing dispute with the Pentagon, where the company has stressed that human oversight is essential for autonomous weapons because current AI models aren't reliable enough to operate without it.

Anthropic didn't win the contract. Instead, the Pentagon awarded it to SpaceX/xAI and two defense companies partnered with OpenAI.

Comment Source: Bloomberg

Maximilian Schreiner

Mar 2, 2026

Short News

ASML, the world's sole manufacturer of EUV lithography machines used to produce advanced chips, is looking to expand beyond its core business. That's according to a Reuters report citing ASML Chief Technology Officer Marco Pieters.

The Dutch company is specifically planning to move into advanced packaging - a technique where multiple specialized chips are connected and stacked on top of each other. This approach is critical for modern AI chips and the high-bandwidth memory that feeds them. TSMC already uses advanced packaging to build Nvidia's most powerful AI processors, among others.

Pieters told Reuters that ASML is planning 10 to 15 years ahead, studying what kinds of machines the industry will need for packaging and bonding. The company is also exploring whether chips can be printed beyond their current size limit. On top of that, ASML wants to use AI to speed up the control software running its machines and improve quality checks during chip manufacturing.

Comment Source: Reuters

Matthias Bastian

Mar 1, 2026

Short News

Artificial Analysis has released version 2.0 of its AA-WER speech-to-text benchmark. ElevenLabs' Scribe v2 leads with a word error rate of just 2.3 percent, followed by Google's Gemini 3 Pro (2.9%) and Mistral's Voxtral Small (3.0%). Google's Gemini 3 Flash (3.1%) and ElevenLabs' older Scribe v1 (3.2%) are close behind. Notably, Google didn't specifically train for transcription—the strong results come from Gemini's general multimodal capabilities. OpenAI's popular open-source Whisper Large v3 (4.2%) lands mid-pack, while Alibaba's Qwen3 ASR Flash (5.9%), Amazon's Nova 2 Omni (6.0%), and Rev AI (6.1%) bring up the rear.

Bar chart showing the AA-WER v2.0 overall ranking with word error rates ranging from 2.3% (Scribe v2) to 6.1% (Rev AI). — ElevenLabs' Scribe v2 tops the AA-WER v2.0 overall ranking with the lowest word error rate, followed by Google's Gemini 3 Pro and Mistral's Voxtral Small. | Image: Artificial Analysis

The results hold up in the separate AA-AgentTalk test for speech directed at voice assistants: Scribe v2 (1.6%) and Gemini 3 Pro (1.7%) pull well ahead, with AssemblyAI's Universal-3 Pro taking third at 2.3%.

Bar chart showing the AA-AgentTalk ranking with word error rates ranging from 1.6% (Scribe v2) to 6.1% (Rev AI). — ElevenLabs' Scribe v2 and Google's Gemini 3 Pro also dominate the AA-AgentTalk voice assistant test with the lowest error rates. | Image: Artificial Analysi

Comment Source: via X

Matthias Bastian

Feb 28, 2026

Short News

The latest generation of large language models—from GPT-5 onward—still struggles when tasks are spread across multiple conversation turns. Researcher Philippe Laban and his team tested current models on six tasks covering code, databases, actions, data-to-text, math, and summarization. Performance drops significantly when information is split across several messages (sharded) instead of a single prompt (concat).

Newer models did slightly better—performance degradation shrank from 39 to 33 percent—but the issue is far from solved. The biggest gains showed up in Python tasks, where some models only lost 10 to 20 percent. Laban suspects real-world losses could be even worse, since the tests used simple user simulations. Users who change their mind mid-conversation would likely cause steeper drops.

Technical tweaks like lowering temperature values don't fix the problem, the original study found. The researchers recommend starting a fresh conversation when things go sideways, ideally by having the model summarize all requests first and using that summary as the starting point for a new chat.

Comment Source: via X

Matthias Bastian

Feb 28, 2026

Short News

In a letter to AI Minister Evan Solomon, OpenAI has promised the Canadian government it will tighten its safety protocols. The move follows a fatal shooting at a school in Tumbler Ridge, British Columbia, that killed eight people. The suspect, Jesse Van Rootselaar, had previously interacted with ChatGPT. An internal algorithm flagged the interactions as possible warnings of real-world violence, and OpenAI employees reviewed them. The company blocked the account but ultimately decided not to contact police.

According to the Wall Street Journal, OpenAI now plans to adopt more flexible criteria for sharing account data with authorities, establish direct lines of communication with Canadian law enforcement, and improve its systems for detecting evasion tactics. OpenAI Vice President Ann O'Leary said the account would have been reported under the new rules. Canada's Justice Minister Sean Fraser warned that new AI regulations could follow if OpenAI doesn't act quickly.

Comment Source: Wall Street Journal

Matthias Bastian

Feb 28, 2026

Short News

Anthropic says it will take the US government to court after Secretary of Defense Pete Hegseth moved to classify the AI company as a supply chain risk, a designation previously reserved for foreign adversaries. Anthropic calls the classification illegal and says it will "challenge any supply chain risk designation in court."

We believe this designation would both be legally unsound and set a dangerous precedent for any American company that negotiates with the government.
Anthropic

Hegseth also implied military suppliers should no longer be allowed to do business with Anthropic. But according to Anthropic, there's no legal basis for that move: the classification under 10 USC 3252 only applies to the use of Claude in direct contracts with the Department of Defense. For private customers, commercial contracts, and access through the API or claude.ai, nothing would change.

The conflict traces back to a failed negotiation process. Anthropic refused to release Claude for mass domestic surveillance and fully autonomous weapons systems, arguing that current AI models are too unreliable for these purposes and that mass surveillance violates fundamental rights. OpenAI has since taken over the deal.

Comment Source: Anthropic

Matthias Bastian

Feb 27, 2026

Short News

Anthropic's dispute with the Pentagon is now rippling through Google and OpenAI. According to the New York Times, more than 100 Google AI employees sent a letter to chief scientist Jeff Dean—who had previously voiced support for Anthropic's position—demanding that Google draw the same red lines: no surveillance of American citizens and no autonomous weapons without human oversight through Gemini. Separately, nearly 50 OpenAI and 175 Google employees published an open letter criticizing the Pentagon's negotiating tactics.

We hope our leaders will put aside their differences and stand together to continue to refuse the Department of War's current demands for permission to use our models for domestic mass surveillance and autonomously killing people without human oversight.
Quote from the open letter "We will not be divided"

According to the Wall Street Journal, OpenAI CEO Sam Altman told his employees that OpenAI is working on its own Pentagon contract that would include the same safety guidelines Anthropic is pushing for. Altman hopes to find a solution that works for other AI companies as well.

Comment Source: New York Times | WSJ | Open Letter

Matthias Bastian

Feb 27, 2026

Short News

Meta has signed a multi-year, multi-billion dollar contract with Google to rent its AI chips—Tensor Processing Units (TPUs)—for developing new AI models. That's according to The Information. Meta is also looking into buying TPUs outright for its own data centers starting next year.

The deal takes direct aim at Nvidia, which dominates the AI chip market and has been Meta's go-to GPU supplier for AI training. Just days earlier, Meta had announced plans to buy millions of GPUs from Nvidia and AMD. Internally, Google Cloud executives have set a goal of capturing up to ten percent of Nvidia's annual revenue—roughly $200 billion—through TPU sales. Google has also launched a joint venture with an investment firm to lease TPUs to other customers.

Here's where it gets complicated: Google itself is one of Nvidia's biggest customers, since cloud customers still expect access to GPU servers. So Google has to keep buying Nvidia's latest chips to stay competitive in the cloud market, while simultaneously trying to eat into Nvidia's market share with its own silicon. OpenAI reportedly managed to negotiate 30 percent lower prices from Nvidia simply because TPUs exist as an alternative.

Comment Source: The Information

Jonathan Kemper

Feb 27, 2026

Short News

A new integration links Figma's design platform directly with OpenAI's Codex. Teams can automatically generate editable Figma designs from code and convert designs into working code. It runs on the open MCP standard, supports Figma Design, Figma Make, and FigJam, and is set up in the Codex desktop app for macOS.

Until now, moving between Figma and code was mostly a one-way street. Dev Mode offered basic HTML/CSS snippets, plugins exported designs as React or HTML, and Figma Make generated React components from text input. These tools worked in isolation without understanding the full project. The new integration creates an end-to-end connection where the AI accesses code, Figma files, and the design system simultaneously.

Figma was one of the first partners with its own ChatGPT app and uses ChatGPT Enterprise internally. According to OpenAI, over one million people access Codex weekly, with usage up more than 400 percent since the start of the year.

Comment Source: OpenAI | Developer Blog