Anthropic launches Code Review for all those bothersome Claude pull requests

Drowning in pull requests from Claude Code? Anthropic has an answer. (Picture: Anthropic)
If you ever used Claude Code, you’d probably notice a mountain of pull requests asking for review on any decent code base. This takes time and effort for developers — but now Anthropic offers a solution.

— One of the questions that we keep getting from enterprise leaders is: Now that Claude Code is putting up a bunch of pull requests, how do I make sure that those get reviewed in an efficient manner? Cat Wu, Anthropic’s head of product, tells TechCrunch.

The answer is the newly launched Code Review tool that uses multiple agents to scan code changes, comment, and rate them for severity.

Using it internally, Anthropic found something of note in 84% of automated code reviews with more than 1,000 lines, they say.

The only problem is that it takes quite a few tokens to run a lot of agents on code changes, and the average cost is between $15-$25 per pull request, depending on complexity, Anthropic writes.

The tool is available as a research preview for Team and Enterprise plans as of today.

Read more: Anthropic’s announcement, writeups on TechCrunch and The Register.

Anthropic’s Cowork gets tapped by Microsoft Copilot

Microsoft’s version of Copilot is entirely cloud based, pleasing many enterprise customers. (Picture: Microsoft)
While announcing Claude availability in «mainline Copilot chats,» Microsoft is also launching its version of the Cowork agent.

Copilot Cowork works mainly across Microsoft’s own Office 365-offerings, and can author emails in Outlook, manage calendars and create files in PowerPoint, Word and Excel, Microsoft says.

— When you hand off a task to Cowork, it turns your request into a plan. The plan continues in the background, with clear checkpoints so you can confirm progress, make changes, or pause execution at any time, they write.

Anthropic’s Cowork does much of the same for general office workloads, and is available as a separate app — with connectors for popular services and tasks.

It works locally on your computer to complete tasks, but Microsoft’s version is cloud-based, and works the data companies have uploaded, offering more diversity and control of their data.

Read more: Microsoft’s presentation and blog post. Writeups on Reuters, The Register and VentureBeat.

Alibaba’s new Rome agent got caught mining crypto without authorization

Mining for crypto was not exactly in the brief for the coding agent during training. (Picture: generated)
In training, Alibaba’s new coding assistant agent opened up an SSH-tunnel to an outside computer and started secretly mining cryptocurrency.

It was an «unanticipated — and operationally consequential — class of unsafe behavior that arose without any explicit instruction and, more troublingly, outside the bounds of the intended sandbox,» the researchers write in a paper, buried on page 15.

It was not prompted in any way to do this, or acting on any kind of instruction — the agent was acting autonomously.

Crypto mining opens up a path to interacting with the real world economy, Axios points out, and lets a rogue agent set up its own business and pay for services, for example.

The actions of the bot was caught by the team’s firewall, along with other attempts to access their «internal network resources,» and was quickly stopped after a brief investigation.

Many are pointing to this incident as a «paperclip»-moment and considering the vast GPU resources that an agent in training has access to, it poses some real questions about agentic security.

Read more: Paper on xArchiv, reports on Axios, Cryptopolitan, and Copilot summary.

OpenAI puts «adult mode» on the back burner, for now

«Adult mode» on ChatGPT gets another postponement. (Picture: generated)
The much touted feature last showed up as «naughty chats» in the Android app just two weeks ago, but is now postponed indefinitely.

It was supposed to debut in the first quarter of 2026, after first being conceived as «letting adults be adults» in October 2025. It was supposed to run in tandem with age verification, rolled out this January.

OpenAI says that «we still believe in the principle of treating adults like adults, but getting the experience right will take more time,» writes Axios.

The postponement is due to the company focusing on other high priority issues, such as «intelligence, personality improvements, personalization,» and «making the experience more proactive.»

Read more: Axios and TechCrunch.

Claude finds 22 security vulnerabilities in the latest version of Firefox

Claude spent two weeks finding a fifth of all serious bugs in all of 2025. (Picture: Adobe)
14 of the bugs Opus 4.6 discovered were classified as «high-severity vulnerabilities» and were fixed by Mozilla in the latest update in late February.

The process took only two weeks to find about a fifth of the total high-severity risks found in all of 2025 — providing a much faster way to scan for bugs.

— Opus 4.6 is currently far better at identifying and fixing vulnerabilities than at exploiting them. This gives defenders the advantage, Anthropic writes, but warns this might change.

Claude works on the full stack, from initial bug hunting to verification and then suggesting patches, offering much needed relief to overworked developers.

— We view this as clear evidence that large-scale, AI-assisted analysis is a powerful new addition in security engineers’ toolbox, Mozilla says in a blog post.

Read more: Anthropic’s workthrough, Mozilla’s blog. Writeups on TechCrunch and Axios.

OpenAI’s robotics and hardware head leaves in Pentagon protest

Caitlin Kalinowski, head of robotics and consumer hardware at OpenAI, makes no qualms about her feelings on their new contract with the Pentagon, and is resigning in protest, saying that:

— Surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation than they got.

Continue reading “OpenAI’s robotics and hardware head leaves in Pentagon protest”

WhatsApp tentatively allows AI chatbots competing with Meta in Europe

WhatsApp sets steep prices for rival AI access. (Picture: generated)
As the EU Commission is considering «interim measures» against the messaging app for refusing chatbots not made by Meta, WhatsApp is slightly opening the door to rivals in Europe.

The platform has 3 billion users, and is considered a «gatekeeper» in EU laws, subject to demands for equal access.

The compromise Meta is rolling out is that rival chatbots will be allowed on the platform, but have to pay their way.

The fees range from €0.0490 to €0.1323 for «non-template messages.» That could ratchet up quickly, considering that chatbot sessions cover multiple messages across millions of users, writes TechCrunch.

The European Commission is said to be «analyzing» how this move «might affect its interim measures» as well as the broader investigation, Reuters reports.

Read more: Reuters and TechCrunch.

OpenAI’s new ChatGPT-5.4 has native computer use and less hallucinations

The latest version of ChatGPT sees a marked jump in the benchmarks. (Picture: Adobe)
The new Thinking and Pro models are more «capable and efficient» and is the first OpenAI model with native computer use skills. It also improves on hallucinations and Office files creation — areas where Anthropic has been thriving.

— Together with advances in general reasoning, coding, and professional knowledge work, GPT‑5.4 enables more reliable agents, faster developer workflows, and higher-quality outputs across ChatGPT, the API, and Codex, OpenAI writes.

On hallucinations, it is 33% less likely to be wrong in its responses, and 18% less apt to have mistakes in replies compared to GPT‑5.2.

Continue reading “OpenAI’s new ChatGPT-5.4 has native computer use and less hallucinations”

OpenAI ships Codex app for Windows — running in the Linux subsystem

The official Codex app has finally arrived for Windows, almost a month to the day after first debuting on macOS.

It runs natively in the Windows Subsystem for Linux, and has integrated terminals for PowerShell, Command Prompt, and Git Bash.

It also has all the «regular» features the Mac version has — and is a fully integrated multi-agent coding environment.

The Windows app is also properly sandboxed, so you can block it from accessing files outside your working folder and «prevent outbound links» unless you approve it, writes OpenAI’s Andrew Ambrosino on x.com.

In addition, your session history is stored on your OpenAI account, making it possible to start coding on a Mac and finish it on Windows without losing work, Engadget notes.

The app is available today from the Microsoft Store.

Read more: Andrew Ambrosino’s announcement, Engadget and Microsoft Store.

NotebookLM introduces «cinematic video overviews» feature

The AI learning and note-taking app can now illustrate your research through «rich, detailed visuals» in full bore video.

Previously, it could only make a slide show of your notes, in addition to the killer feature of creating podcasts from them.

The new videos are possible through a combination of Gemini 3, Nano Banana Pro and Veo 3 — with Gemini «acting as a creative director.»

The Gemini model makes «structural and stylistic» decisions on the fly, illustrating content word by word and in context to create something like the video above.

The feature is only available through the $300/month Google AI Ultra subscription, but it is sure to trikle down at a later stage

Read more: Google’s announcement, Android Police and The Verge.

Jensen Huang says Nvidia’s investment opportunity in AI labs is closing

Huang figures the privately owned AI labs era might be finished. (Picture: Nvidia)
The Nvidia CEO says the opportunity to invest might soon end, Reuters reports.

The reason for this is straightforward, suspecting that Anthropic and OpenAI going public «later this year» will shutter the window to private equity deals.

The latest deal to fund OpenAI with $30 billion «might be the last time» to «invest in a consequential company like this,» Huang admits.

Nvidia has invested some $130 billion in OpenAI in two rounds, the recent straight up investment, and one circular deal where they paid $100 billion in return for OpenAI buying $100 billion in chips from them.

Likewise, Nvidia was an investor in a November funding round for Anthropic, buying $15 billion in shares from the company.

Read more: Reuters, CNBC and TechCrunch.

OpenAI eyes unclassified NATO contract, The Wall Street Journal reports

OpenAI seems to be making a big push for defense contracts. (Picture: DOD photo by Navy Petty Officer 1st Class Dominique A. Pineiro, CC BY 2.0)
Just days after striking a deal with the Pentagon, it would seem that OpenAI is actively pursuing more deals in the defense sector, writes The Wall Street Journal.

During an all-hands meeting at OpenAI on Tuesday, Sam Altman tried to clear the air on their recent Pentagon contract debacle — which has seen some twists and turns before landing on very clear language against use in mass surveillance.

Of note at the meeting, he said that OpenAI was considering a contract to deploy on NATO’s classified network, although he was confirmed to have misspoken by a spokesperson, saying the deal being considered was for the unclassified parts of the defense alliance.

Later on Wednesday, Reuters confirmed that such a deal is indeed under consideration, citing sources «familiar with the matter.»

While talking, he said, among other things, that the US military had been a «great benefit to all of humanity over the last 250 years,» The WSJ writes, and added that «Clearly, the military has done things that I extremely disagree with, and am sure will do more in the future.»

Read more: The Wall Street Journal and Reuters.

OpenAI launches GPT-5.3-instant with fewer lectures and less hallucinations

GPT-5.3-Instant performs better than 5.1-Instant and is a little behind 5.2-Instant, but is better at conversation. (Picture: Adobe)
Hallucinations are down 27% when it uses the web and ~20% in reasoning, and OpenAI did actually notice that the 5.2 model often cames across like a nanny — berating users with obvious, condescending lectures for harmless questions.

This has been solved by reducing unnecessary «dead ends,» caveats and «declarative formulations» in order to let the conversation flow more freely — solving one of many users’ major gripes with the current OpenAI models.

The model is also better at search, combining search responses with its own knowledge and reasoning. This should put results into context instead of just summarizing results.

Bu the real update with this model is bringing ChatGPT to a more natural style and tone with freely flowing conversation — which should be far less «cringe.»

OpenAI is promising less «Stop. Take a breath»-responses, or «I’m going to calm this down» kinds of outputs. It should also be better at creative writing.

GPT-5.3-instant is available today in the app (choose Instant in the model selector), API and web, and updates to Thinking and Pro are «coming soon.»

Read more: OpenAI’s announcement, writeups on 9to5Google, TechCrunch and The Register.

Google debuts Gemini 3.1-Flash Lite, for developers needing speed and scale

Not for everyone; Flash Lite is built for high volume cost efficiency. (Picture: Google)
Positioning the model as a purely developer-focused one, Google is touting the price, latency and the sheer amount of work it can do.

Costing $0.25 for 1M input tokens and $1.50 for 1M output tokens, it is one of the cheapest models out there.

Compared to Gemini 2.5 Flash, it is 2.5x faster to the first answer, and 45% quicker in output speed, while maintaining quality.

This benefits high-frequency workloads, such as mass translations and content moderation where price is a priority, Google says.

Users of AI Studio and Vertex AI can also adjust its thinking levels, making it possible to balance speed and complexity.

Read more: Google’s announcement, Android Central, Tom’s Guide.

U.S. Supreme Court declines to hear AI copyright case, making it final

AI art is not covered by Copyright, the courts have decided. Cutout of «A Recent Entrance to Paradise,» by Stephen Thaler’s AI.
Computer scientist Stephen Thaler’s quest to have his AI outputs protected by copyright has come to an end, Reuters reports.

The lesser courts had consistently upheld the Copyright Office’s 2022 decision to deny his picture «A Recent Entrance to Paradise» protections, holding that Copyright must have clear human authorship.

The picture was created solely by his AI system «DABUS» in 2018, and Thaler said that «The Copyright Office ​will have irreversibly and negatively impacted AI development and use in the creative ​industry during ⁠critically important years.»

The case has wide repercussions for the creative industry’s use of AI in artworks, but yet to be covered by the courts are several cases with significant human involvement, where the AI assists human creativity instead of creating it alone.

Read more: Reuters, The Verge.