Malicious Claude Code Extension Hits Cursor as AI Companies Unveil New Tools

Friday, 30 October 2026 at 19:48

A malicious version of the Claude Code extension was distributed through Cursor's marketplace on October 29, potentially exposing up to 5,000 users. Today's AI World Today: Daily News also covers Mistral's downloadable AI model, Canva's new editing assistant and Amazon's soaring infrastructure spending.

The October 30 episode examines a particularly busy period for developers. Security researchers are investigating how a malicious extension reached users, while AI companies are introducing new ways to run models, edit designs and control the cost of coding assistance.

We finish with an unusual project that brings modern AI to old Nokia phones.

Listen to AI World Today: Daily News

Listen on Spotify, Apple Podcasts or Amazon Music.

Cursor warns users about a malicious Claude Code extension

Cursor published a security notice after discovering that a modified version of Anthropic's Claude Code extension had been distributed through its marketplace.

The affected version, 2.1.313, was available between approximately 2:27 a.m. and 1:35 p.m. Pacific Time on October 29.

Cursor estimates that at most 5,000 users were affected. That number represents potential exposure, not confirmed successful compromises.

According to the company, the malicious code collected information about affected computers, installed an unauthorized SSH key and attempted to establish persistent remote access.

The attacker-controlled server has since been taken offline. Cursor also removed the extension, temporarily shut down its marketplace, rotated security keys and rebuilt the service.

The company says it has no evidence that other versions or the official distribution were affected. The modified version was not developed by Anthropic.

For potentially affected users, uninstalling the extension alone may not be sufficient. Cursor has published checks and remediation scripts for Windows and macOS, alongside guidance on reviewing credentials.

The incident illustrates a wider software security problem: even when developers trust an AI assistant, the extensions and other tools used to access it introduce additional risks.

Mistral releases the weights of Large 4

French AI company Mistral has released the model weights of Mistral Large 4, nicknamed Le Chonk.

The release follows the model's public preview earlier in October. Mistral says additional training has improved coding performance, defensive cybersecurity capabilities and reasoning efficiency.

Large 4 contains approximately one trillion parameters in total, with 52 billion active parameters. It offers three reasoning levels, allowing developers to balance task complexity against computing costs.

The most significant change is deployment flexibility. Organizations can download the weights and run the model on infrastructure they control rather than relying exclusively on a hosted API.

That is particularly relevant for businesses and public institutions with strict data residency, customization or operational requirements.

There are important qualifications. Running a model this large requires substantial computing resources. The release also uses Mistral's Ecosystem Open-Weight License, which requires a separate commercial license for businesses offering the model itself as a service.

Its reported benchmark improvements are promising, but organizations will need to test performance on their own workloads.

Canva Brush edits designs instead of rebuilding them

Canva has introduced Brush, a vision-language model intended to understand the structure of existing designs and make requested changes.

The distinction from conventional image generation is important. A finished presentation contains separate layers, text elements, images, colors and layout decisions. Editing one part without disturbing the rest can be difficult.

Brush is designed to work within that structure.

For example, a user could request a different visual style across several presentation slides or adapt an existing design to another format. The resulting elements remain editable inside Canva.

Canva reports that its internal testing found editing tasks up to 69% faster and 46% cheaper to run than with its previous design agent. Those are company-reported figures rather than independently verified results.

Brush is initially available for supported editing tasks to Canva Pro subscribers, with broader access planned.

For businesses creating regular marketing material, this development could make revisions more convenient. The important test will be whether the AI preserves the details users did not ask it to change.

Amazon's cloud growth comes with an enormous infrastructure bill

Amazon Web Services reported $46 billion in third-quarter revenue, up 39% from the same period a year earlier.

The result marked AWS's fastest growth in nearly five years. The division also reported operating income of $17.8 billion.

Amazon's quarterly earnings release shows how quickly cloud computing demand is growing.

It also shows the scale of the accompanying investment.

Amazon reported negative free cash flow of $42.4 billion over the trailing twelve months. That is not a single-quarter figure, nor does it mean the company reported an accounting loss.

For the third quarter, Amazon generated $28.1 billion in operating income across its businesses.

The decline in free cash flow largely reflects increased spending on property and equipment, which Amazon says was primarily driven by AI investment.

The company still expects approximately $220 billion in cash capital expenditures during 2026.

That creates a financial challenge shared by several major technology companies. Demand for AI computing can be exceptionally strong, while supplying it requires enormous expenditure before the equipment begins generating revenue.

GitHub Copilot Fusion chooses models for coding tasks

GitHub Copilot Fusion is now generally available.

Instead of requiring developers to select an individual AI model, Fusion can choose a model or coordinate multiple models for a particular task.

Its approaches include using a single model, escalating difficult tasks to more capable systems, or having one model review another model's output.

Developers choose between Standard, High and Max effort.

GitHub says its highest setting performed competitively with leading coding agents in benchmark testing while reducing average token usage, execution time and API costs.

Actual costs depend on the models selected. GitHub bills for the models used and is offering a temporary discount on Fusion model costs for paid subscribers.

For developers, the practical attraction is that model selection becomes part of the tool's work rather than another decision they must constantly manage.

AIKON brings modern AI to old Nokia phones

The final story is a smaller, more playful experiment.

Developer Emir Karşıyakalı has created AIKON, an application that brings access to modern AI services to certain older Nokia phones.

The project supports Java-capable devices and has been tested on the Nokia 6300 and E63.

The phones do not run large language models themselves. Instead, an application on the device communicates with a separate server, which sends requests to AI providers and returns the answers.

AIKON supports services including Claude, ChatGPT, Gemini and Grok. The source code and technical setup guide are publicly available.

Older hardware still comes with limitations, including small screens, restricted input methods and network compatibility issues.

Nevertheless, the project demonstrates something interesting: using modern AI does not always require a modern device, provided the device can communicate with a service capable of doing the processing elsewhere.

Follow AI World Today: Daily News

AI World Today: Daily News covers the latest developments in artificial intelligence, from major technology companies to independent experiments and practical applications.

Follow the show on Spotify or Apple Podcasts, or your favourite podcast app!