Created automatically with AI: the texts and images in this edition are AI-generated, based on the linked original sources. Errors may occur.
Apple wants to strengthen safeguards around apps’ full access to files on the Mac. There are also new initiatives for AI hardware and workforce training, as well as debates about safety, open models and the regulation of manipulated real-estate images.
AI-generated image
01heise online
Apple wants stricter safeguards for full file access on the Mac
Apple has announced that it will more closely tie the macOS “Full Disk Access” permission to users’ explicit consent in the future. It allows apps to access all files, including messages, emails and browsing history. When and how the change will be implemented remains unclear. Apple cites the growing risks posed by more capable and autonomous AI agents as the reason for the move.
Why it matters
The permission gives apps access to numerous private files. In light of autonomous AI agents, Apple wants to strengthen users’ control over that access.
Meta opens Muse Gadgets to custom-built AI devices
Meta has introduced Muse Gadgets as an open-source project. Hobbyists can use it to connect devices based on ESP32 boards or Linux systems to the Muse AI agent. The provider supplies firmware and a Linux SDK. Meta has also produced 5,000 Muse Home Links, according to the company: USB-C devices through which Muse can communicate with a home network and compatible smart-home devices. They are to be offered free of charge to Muse subscribers.
Why it matters
The project expands access to Muse beyond apps to include custom-built devices and connected household appliances. Meta says it provides both firmware and a Linux SDK.
OpenAI safety staffer leaves company and criticizes its culture
David Robinson, who according to TechCrunch worked on safety reports about OpenAI products, is leaving the company and criticizing its safety culture in an opinion piece. He argues that a trial-and-error approach could lead to bigger mistakes with more capable systems, and calls for multilayered safeguards. OpenAI said it was improving testing, monitoring and safety measures, among other things.
Why it matters
The departure adds to a public debate about safety standards at leading AI companies. Robinson’s assessment comes from a former employee; OpenAI points to ongoing improvements.
Anthropic announces training program for 10,000 AI professionals
Anthropic is launching the Claude Frontier Academy and says it plans to invest $100 million to train a total of 10,000 so-called Frontier Deployed Engineers by the end of 2027. The program combines in-person training with a 12-week practical phase on an AI project at each participant’s company. According to the provider, the first participants include people from consulting firms and banks, among others. Organizations nominate participants.
Why it matters
The initiative is intended to qualify professionals to introduce AI systems in companies. The scope, funding and target number are figures provided by the company; Anthropic expects the first completion awards to be presented in early 2027.
According to The Batch, Xiaomi has released MiMo-V2.6-Pro-RL, along with a smaller variant and a distilled model with open weights. The report cites good results for the Pro model in two independent overall evaluations, but also notes slower outputs and a gap behind leading models. Xiaomi also provided more than 7,000 training environments and the training code, according to the report. Performance and cost comparisons depend on the tests in question.
Why it matters
Open weights and published training materials can make independent evaluation and further development easier. The reported evaluations do not establish performance across all applications.
Australian authority mistakes real mural for manipulated real-estate photo
A law in New South Wales aims to protect against misleading depictions in real-estate listings by addressing edited images. According to the report, a listing was removed because a real, large-scale mural was mistaken for a digital alteration. The landlords had to prove that the artwork actually existed. The incident highlights a practical difficulty in applying the rule: a genuine feature was initially classified as an altered image.
Why it matters
The case shows that enforcing a rule against manipulated real-estate images can lead to mistakes. Here, a real mural was wrongly mistaken for a digital alteration.
Automatically summarized with AI and checked against the source texts retrieved. Original sources are linked for every story; errors may still occur. An editorial overview from Medienfuchs.
