Anthropic uses Claude to write new Claude «constitution»

Claude’s new constitution is written in part by asking Claude. (Picture: Anthropic)
Claude has gotten a new constitution, written in part with the help from previous versions of Claude — and it marks a change in Anthropic’s approach

— While writing the constitution, we sought feedback from various external experts (as well as asking for input from prior iterations of Claude), Anthropic says.

The new constitution is going to tell Claude how to behave in broader, more ethical terms, they write.

This is a departure from previous constitutions that were big long lists of specific principles and interactions, that detailed how Claude would act.

The bot needs to generalize more to decide on situations not predicted in the written guide, Anthropic says.

The constitution for Claude is the «foundational document» for how the bot should act, and is used in both training and inference (as in day-to-day use). It is supposed to be a living document, getting updated continuously as Anthropic sees how the bot behaves.

Read more: Anthropic’s announcement, the actual Constitution. Writeups on TechCrunch, Time.com, Axios.

Ukraine says to share valuable wartime drone data with allies to train AI

A Valkyrie AI drone flies with an F-22. Now, they might get upgrades. (Picture: Tech. Sgt. James Cason, USAF)
Ukraine’s new defense minister, Mykhailo Fedorov, is sitting on a trove of metrics from four years of drone warfare — and is now using it to train AI.

They will now «establish a system allowing its allies to train their artificial intelligence» on it, Reuters reports.

They are already launching a project with Palantir to produce an AI on the combat data.

The data includes «extensive battlefield information, logs of combat statistics and millions of hours of drone footage,» Reuters writes.

Training an AI for combat needs tons of real-world data, preferably from a real battlefield, not simulations — and this data will be immensely important for Ukraine’s allies, who are all busy developing combat AI drones as we speak.

Read more: Reuters, Financial Times and The Independent.

First new AI models from Meta released internally this month

Meta’s Avocado is progressing, but there’s a “tremendous amount of work” still to do. (Picture: generated)
Meta’s Superintelligence Labs have released some models internally, CTO Andrew Bosworth tells Reuters.

The models «show a lot of promise» and are «very good,» he says.

Meta has been rumored for a while to have a couple of models in the works, including the Avocado model for text and an image/video-focused model named Mango.

These models now seem out of training are showing promise internally.

Bosworth also warns that there’s «a tremendous amount of work to do post-training.»

Meta hasn’t launched a new model since Llama 4 in April 2025, but Bosworth is hinting to Reuters that 2026-27 are «important years for bringing consumer products to market.»

Read the full scoop at Reuters.

It’s official: OpenAI rolls out age-gating globally

OpenAI will use AI to verify your age and serve a less harmful experience, but it does make mistakes. (Picture: generated)
Last month offered a sneak peek at OpenAI’s support page for «age prediction,» and now they are ready to go live.

Age prediction works by taking in several factors such as usage time, account age and the time the user is active.

It is developed «in dialogue» with experts from the American Psychological Association, ConnectSafely, and Global Physicians Network⁠.

After several wrongful death lawsuits involving teens, disturbing mental health conversations and stating that principles are in conflict, OpenAI was more or less forced into doing something on both teen use and parental controls.

Those deemed under 18 by the AI will be served a vanilla version of ChatGPT without beauty standards, violence or romantic role-play, to name a few examples.

If someone gets wrongfully lobbed into the teen experience, and it does make mistakes, they can upload a selfie for proper verification through Persona.

Read more: OpenAI’s announcement, writeups on Reuters, CNBC, Engadget.

OpenAI commits to power buildouts for Stargate centers

OpenAI is now addressing local concerns for their massive data center buildouts.
The company shares how it plans to deal with «Stargate Communities» across the USA, especially on water and power consumption — and commit to being «good neighbors.»

This means for instance that they will cooperate with power utility companies at each site to ensure they are «paying our own way on energy.»

That could entail building out the infrastructure for each site where needed, or simply strengthening the grid.

They also address water usage, and note that Stargate Abilene will use as much water annually as the community uses in a day, thanks to «innovations in the cooling water systems design.»

OpenAI is even committing to slowing down workloads on days with adverse conditions, to lighten the load on the grid.

There have been at least 25 data center cancellations in the USA due to opposition from local communities, Gizmodo reports.

They are typically worried about rising electricity and water costs, which OpenAI is now directly addressing.

Read more: OpenAI’s blog, Reuters, Bloomberg.

Google’s Veo 3.1 hits 179 million Workspace users

Just one of many creative uses of Veo3.1. Jelllyfish in the backseat of a car.
Announced late last week, Google Flow is coming to Workspace.

Flow is a wrapper for AI generated videos, using the popular Veo 3.1 — one of the most advanced generators out there.

The expanded access is for Business and Enterprise (9 million users), and Education plans — which has 170 million users.

These plans can now also access the popular Nano Banana Pro service, letting users generate pictures from anything they can imagine.

The Flow service was previously reserved for Gemini Pro and Ultra users, who could generate 8 second, high quality 4K AI videos in widescreen or vertical formats, and daisy chain them for even longer ones.

The service now dwarves OpenAI’s Sora 2 service, which had 6 million total downloads in January 2026.

Read more: The Verge, and Google’s support document.

OpenAI’s latest numbers; 3x yearly financial/compute growth since 2023

Impressive tally; OpenAI shows unprecedented growth in their new numbers. (Picture: OpenAI)
OpenAI is scaling like never before, according to CFO Sarah Friar, who is out with some hard numbers.

Friar says revenue and compute grow in tandem with the advent of more powerful models — that they «scale with intelligence,» so to speak.

Looking at the numbers, compute has gone from 0.2 gigawatts in 2023 to 1.9 GW in 2025, growing about 3x every year since ChatGPT’s debut.

Financially, OpenAI now has $20 billion in revenue, following the same curve as the compute scale and growing about 3x per year from $2 billion in 2024.

All this is of course before ads arrive on the free tier, and before OpenAI sees any results from its global rollout of the go subscription on their revenue.

On the compute side, OpenAI is chasing infrastructure like there is no tomorrow, closing in on more than 30 GW before 2030 in what would be truly explosive growth — and we can only wonder as to how advanced frontier AI models will get by then.

Read more: OpenAI’s announcement, writeups on Reuters, .

Musk runs into a snag on data center gas turbines — they are now illegal

xAI fits gas turbines on flatbed trucks and calls them «temporary» to avoid regulation. No more, says the EPA. (Picture: generated)
Apparently, a new EPA rule on stationary and «temporary» gas turbines for energy generation has made them illegal, according to CNBC.

They produce much too high levels of nitrogen oxides, and must be regulated as combustion engines, the EPA says.

That means Musk’s and xAI’s Colossus plant will have to rethink their energy use, as they make widespread use of natural gas turbines to generate electricity for their facilities.

They will basically have to get Clean Air Act permits, and prove they aren’t harmful.

The local population have long been complaining of a rotten-egg-like stench in the atmosphere, CNBC writes, and smog is supposedly prevalent.

xAI uses 15 turbines for Colossus 1 and at one point had 59 turbines for Colossus 2, The Guardian writes.

Read more: The EPA Rule. Writeups on CNBC, Gizmodo and The Guardian.

Demis Hassabis: China is just «months» behind US in AI capabilities

China is great at playing catch-up, but can they innovate? That’s the next challenge, Hassabis, says. (Picture: Wikipedia, CC BY-SA 4.0
The CEO of Google’s DeepMind has some choice words for China in a recent podcast.

They might be «just a matter of months» begin Western capabilities, he tells CNBC.

— The question is, can they innovate something new beyond the frontier? So I think they’ve shown they can catch up … and be very close to the frontier … But can they actually innovate something new, like a new transformer … that gets beyond the frontier? I don’t think that’s been shown yet, he tells the new podcast The Tech Download.

The key tech to unlock Chinese AI is access to chips, he says, where the USA is far ahead. The US recently okay’ed exports of the powerful H200 chip from Nvidia, but reception in China has been lukewarm from authorities.

— To invent something is about 100 times harder than it is to copy it, says Hassabis on the podcast. — That’s the next frontier really, and I haven’t seen evidence of that yet, but it’s very difficult.

Read the full scoop on CNBC.

OpenAI to start «public» advertising tests on ChatGPT

What ads will look like If you click it, you can query it for more information. (Picture: OpenAI)
The tests will happen «during the coming weeks» for ChatGPT Free and Go tiers — and will try to put relevant, clearly labeled ads beneath GPT responses, OpenAI announces.

OpenAI says they won’t share your data or conversation to advertisers, and will «maintain a high standard» where you can turn off personalization if you want to.

They won’t be shown to under-18s, they say, nor will they show on sensitive topics, such as physical health, mental health or politics.

The neat part of the coming ads is that you click on them and query them for further detail, which is a feature not found in traditional advertising.

The Financial times estimates that OpenAI can earn somewhere around the «low billions» from advertising.

Ads won’t be shown for paying tiers, such as Plus, Pro and Enterprise.

Read more: OpenAI’s announcement. Writeups on Reuters, The Verge and Ars Technica.

ChatGPT Go expands globally, comes to the United States

A little more dollars for a little more usage. It’s a simple plan, yet effective. (Picture: generated).
ChatGPT Go was first launched as a cheap way to access more GPT queries in India in August, 2025, for around $5. Now, the subscription has become popular enough to expand globally to every market where OpenAI is active.

The idea is that for a just little more money, $8 in the USA, you can get ten times the queries, ten times the file use, and ten times the image generations compared to the free plan.

OpenAI says that since the introduction of the Go subscription tier, they have seen extensive everyday use in tasks such as writing, learning, image generations and problem-solving.

Along with the Free tier, ChatGPT Go will be showing ads once they are ready.

Read more: OpenAI’s announcement, OpenAI’s sub overview. Writeups on The Verge, MacRumors.

Gemini Pro and Ultra subscribers get upgraded, decoupled usage limits

Pro and Thinking modes of Gemini 3 no longer draw from the same usage allocation. (Picture: Generated)
Using the Thinking and Pro Gemini models used to draw from a shared pool of a hundred available prompts — and this is no longer the case.

Now you get separate usage limits for the models, so you no longer drain your Pro pool by using the Thinking model or vice versa.

At the same time, Google is upgrading the total limits for the Thinking model to up to 300 prompts per day for the Pro plan, while leaving the Pro model allocation at 100 per day.

Ultra subscribers get 500 daily prompts on the Pro model and 1,500 prompts on the Thinking model.

The upgrade is due to user feedback, writes 9to5Google, quoting Google:

— Many of you want more precision and transparency when deciding which model to use for your daily tasks.

Read more: Google: Rate limits on Gemini, writeups on 9to5Google, Mashable.

Amazon, Meta, Microsoft, Mistral and Perplexity join Wikipedia’s AI program

Some of the companies joined last year, but this is the first time it’s made public. (Picture: Wikimedia, CC BY-SA 4.0)
The Wikimedia Enterprise collaboration grew by five new members in time for Wikipedia’s 25th anniversary, joining to receive structured data directly from the web’s largest knowledge repository.

— In the AI era, Wikipedia’s human-created and curated knowledge has never been more valuable, Wikimedia, the encyclopedia’s parent, writes in a blog post.

They currently have more than 65 million articles in 300 languages, and get more than 15 billion views a month.

The Wikimedia Enterprise program is both a way to support Wikipedia’s efforts, as it is a paid service, and gives access to some powerful data.

The on-demand API gives labs the most recent version of specific articles, the snapshot API provides a downloadable file for every language and is updated hourly, and the realtime API is a stream of updates as they happen.

Together, they provide better access than just scraping the data off the public servers, and Wikipedia hopes for more to join, if only to stave off having to pay for extensive scanning of its archives — which is more costly.

Read more: Reuters, CNBC, The Verge.

OpenAI to spend $10 billion on compute from AI chip startup Cerebras

Cerebras makes powerful inference chips, for when an AI needs to think a little deeper. (Picture: generated)
The deal will land OpenAI with an added power capacity of 750 megawatts — but it’s not just any kind of compute.

Cerebras makes wafer-scale inference chips (for generating the response an AI gives after a query), looping in networking and high-bandwidth memory on the same die.

This makes for much faster thinking on complex tasks, should enable real-time reasoning, and OpenAI could possibly route those kinds of queries to this kind of compute.

The deal is worth over $10 billion, Reuters writes, and the capacity should come online in multiple tranches to be fully delivered in 2028, OpenAI says.

Sam Altman is an early investor in the company, and OpenAI once considered buying it, TechCrunch reports.

Read more: OpenAI’s announcement, writeups on Reuters, CNBC and TechCrunch.

xAI moves to technically block non-consensual «undressing» edits

Turning real people into pinups is officially over on x.com. (Picture: generated)
xAI’s «Safety»-account has posted a long screed on non-consensual sexualized content delivered through the @Grok account — and it intends to block it:

— We have implemented technological measures to prevent the Grok account from allowing the editing of images of real people in revealing clothing such as bikinis. This restriction applies to all users, including paid subscribers, the post says.

That basically means the fountain is shut off for the controversial «undressing» pictures on the platform for good, and not just limited to paying users. It’s off, period.

It further says that all image generation and edits will only be available to paying subscribers, making it possible to trace any violations to real people.

They also say they are now geoblocking risqué images in countries where such content is illegal.

That’s what concerted international pressure can do. Watchdogs and attorneys general across the world were only just starting their investigations into Grok and xAI, and it had already been blocked in Indonesia and Malaysia.

Read more: Ars Technica, BBC.