In the Oct. 8 episode of 9to5Mac's Security Bite podcast, host Arin Waichulis walks through the new controls Apple told developers it is adding to Full Disk Access — the macOS permission that lets an app read nearly everything on a user's drive. The report ties the move to the rise of AI agents on the Mac, but Apple has not spelled out exactly what will change, and speculation has run high. The episode also explains what Full Disk Access is, what prompted the change, and what remains unknown.
Apple Watch Series 12 went on sale September 18, 2026, starting at $399 in 42mm and 46mm sizes with aluminum, titanium or ceramic cases. It uses an S11 chip with a 4-core Neural Engine, a redesigned Health Sensing System that can take heart-rate readings as often as every five seconds, and a new 0–10 readiness score combining activity, Training Load, Vitals and sleep (requiring at least seven nights of overnight Vitals data). New AI features include Siri AI plus Audio Intelligence — Sound Recognition, Music Recognition, Live Rewind and Siri Recap — limited to Series 12 and Ultra 4; Siri AI is still in beta, needs an Apple Intelligence-capable iPhone and has regional limits.
Well-known tech bull Dan Ives left Wedbush Securities after nearly eight years to launch the newly formed Yorkville Ives, where he restarted tech equity research with an Outperform rating on all 49 stocks he initiated on. He argues AI infrastructure supply will stay constrained at least through 2027 and that the market underestimates a coming wave of as much as $4 trillion in AI spending, likening the AI buildout to the third inning of a baseball game. His top five picks for 2027 are Nvidia, Microsoft, Palantir, Apple and CrowdStrike, with initial price targets of $400 for Alphabet and Apple, $300 for Amazon, $900 for Meta, $615 for Microsoft, $300 for Nvidia, $500 for Tesla and $225 for SpaceX.
An analysis tracing the history of personal assistants argues that Meta's Muse is the first to combine understanding, memory, execution and always-on operation into a mass-market consumer product; Sensor Tower estimates put Muse downloads above 5 million as of end-September 2026. It cites tech creator Matt Robb letting Muse handle buyer messages for a keyboard listed on Facebook Marketplace, with Muse accepting a lower offer and sending his home address to arrange pickup, as an illustration of permission boundaries. The piece threads Siri, Alexa, Google Assistant and OpenClaw together with ByteDance's Doubao and Alibaba's Qwen, and says Apple's redesigned Siri AI in 2026 still centres on understanding what the user is doing on-device rather than delegating long-running tasks.
Analysts led by DA Davidson's Gil Luria raised their Micron price target to $3,000 from $2,100 and reiterated a Buy rating, arguing that AI-driven memory demand will keep the company growing over the next three to five years and that memory will be undersupplied in 2027 and 2028. The report said demand now comes from large US enterprises such as Amazon, Microsoft, Google, Nvidia and Apple, and that longer supply agreements with customers improve visibility. Micron shares rose 4.06% to $1,088 on Oct. 7; the company is expected to start a large buyback in December, when CHIPS Act restrictions on repurchases lapse.
GeekPark's "Creation 100" column (episode 7) rounds up five new hardware devices: the MUZIM L1 desktop data dock, which runs a local OpenSoul file system with "Vibe Search" for natural-language search across photos, video (down to a specific frame) and transcribed audio, plus automatic face grouping and semantic filing. The Lumeria Lumoscope is a four-spectrum skin scope that clips onto a phone camera with an AI coach tracking skin changes, while the Flowtica verso is a MagSafe AI recorder card for the iPhone with a FlowMark highlight button and a press-to-ask Ask AI key. The piece also covers GENiEX, an AI walkie-talkie toy built on a decade-old sci-fi universe, and the immurok IK-1 wireless fingerprint key that gates sensitive commands issued by AI coding agents. Most are still at crowdfunding or pre-order stage, and search accuracy, content depth and manufacturing capacity remain unverified.
Apple Machine Learning Research published Normalizing Trajectory Models (NTM) at NeurIPS, which models each reverse diffusion step as an expressive conditional normalizing flow with exact likelihood training, pairing shallow invertible blocks per step with a deep parallel predictor across the trajectory and trainable from scratch or from pretrained flow-matching models. Its exact trajectory likelihood also enables self-distillation: a lightweight denoiser trained on the score function induced by the model itself produces high-quality samples in four steps. On text-to-image benchmarks, NTM matches or outperforms strong image generation baselines in just four sampling steps while uniquely retaining exact likelihood over the generative trajectory.
Ring founder Jamie Siminoff said the company is expanding beyond its signature video doorbell into a broader smart-home security ecosystem, including a new smart lock. He argued that cameras, locks and other connected devices will become the physical "appendages" of increasingly powerful AI systems inside the home, and addressed rising competition as Apple prepares to enter the category.
Speaking at MIT Future Fest, Tony Fadell — credited with the iPod, the iPhone and as founder of Nest — said the first wave of AI hardware (Rabbit R1, Humane AI Pin, Limitless pendant) failed because it “didn’t meet any kind of need” and was just interesting technology for geeks. He argued trust and safety are paramount for a personal AI assistant, noting Meta’s Muse had a serious vulnerability found soon after launch, and that Apple has the hardware, chips and privacy credibility but no world-class model of its own — the new Siri AI runs on custom versions of Google’s Gemini. He predicted a successful agent will have to run on-device only, for privacy and to stay lightweight.
Google has opened SynthID Detector globally, letting users upload images, videos and audio to check for hidden SynthID watermarks. Engadget and TechRepublic report that the free service supports media from Google, OpenAI, NVIDIA and Kakao, while Google says Apple Intelligence support is coming soon. CNET correctly flagged a ChatGPT-generated image and an image edited with Apple's Reframe tool, but also found that an image generated through Apple's Playground was missed, highlighting the detector's limited coverage.
Rapid-MLX is an Apache-2.0 local LLM inference server built on MLX for Apple Silicon, exposing both OpenAI (/v1/chat/completions, /v1/responses) and Anthropic Messages (/v1/messages) APIs and focusing on reliable tool calling for coding agents. The project claims up to 4× faster decoding than Apple's mlx-lm on identical weights: a 1.50× median across 18 streaming tasks with Qwen3.5-9B 4-bit on an M4 Pro Mac mini, and up to 4.27× on a whole-file code edit (46.4 to 198.4 tok/s), plus 3.0× Ollama's aggregate decode at 8 concurrent streams on Qwen3.6-35B-A3B. All figures are the project's own measurements, and much of its comparison table is drawn from other projects' own documentation.
Apple currently offers controls to restrict individual Apple Intelligence features on Macs, but not a single switch to disable all features and remove downloaded models. 9to5Mac reports that growing opposition to AI, storage concerns, and organizations that ban AI tools could increase pressure on Apple to add an official removal option, though Apple has not announced one.
Google Project Zero says software vendors should prepare emergency mitigation mechanisms for actively exploited vulnerabilities, rather than relying only on faster formal patches. It outlines four approaches: feature flags, input filtering, alternative update channels, and hotpatching, while noting that testing remains necessary.
A report says Hon Hai Technology Group and Foxconn Industrial Internet are benefiting from surging AI-server demand, alongside their established Apple manufacturing business. Foxconn Industrial Internet reported strong 2025 revenue and profit growth, while its cloud-computing business and AI-server sales became major growth drivers, according to the cited financial disclosures.
AI hardware makers are blurring the boundary between sensing and recording by processing audio or video into transcripts and summaries without necessarily saving the raw data. The column argues that these outputs can still preserve information about people, raising unresolved questions about consent, privacy signals and who defines a recording.
Apple Machine Learning Research published RISED, a method that repurposes rubrics to guide online data selection and policy supervision when training a single LLM agent jointly across diverse interactive environments. An LLM judge tags each rollout with a rubric vocabulary shared across environments; the resulting profiles select data that matches the mixed-environment batch's behavioural composition while limiting overlap, positive rubrics provide privileged context for an on-policy self-distillation teacher's token-level supervision, and negative rubrics steer later rollouts away from recurring failure modes. The paper reports that across model backbones RISED achieves the highest mean pass rate across environments and ranks first or second in every individual environment.
CNET explains how to check Apple Intelligence storage usage and disable or limit related features on iPhone and Mac. The steps vary by operating system; changing the Siri language on macOS 27 reportedly freed about 4 GB for the author, while another Mac method did not work in testing.
RemoveMacAI is an open-source command-line tool that removes selected or all Apple Intelligence features from macOS 27 using an approved configuration profile and Apple’s asset service. Its developer says the approach leaves System Integrity Protection enabled and avoids directly modifying /System; users can revert the changes and restore the models when features are re-enabled.
Apple’s redesigned Siri AI can search content from supported apps through App Intents and may process some requests on-device or through Private Cloud Compute. Engadget explains how users can disable app content indexing, stop Siri data sharing for model training, or switch back to Siri Classic.
Apple says future macOS versions will add controls around Full Disk Access, requiring clearer user action before granting the highly privileged permission. The analysis links the change to rising risks from personal AI agents that can read sensitive data and act on a user’s behalf, while noting the tradeoff for legitimate backup, security, and IT-management tools.
Apple Machine Learning Research and Stanford University authors published "Negotiating Ontological Boundaries in User-Authored Personal Sensing Systems." They built two open-ended Wizard-of-Oz probes that let participants experience training a personalized machine learning system on phenomena they define themselves, then ran a week-long exploratory study. The authors identify four sites of ontological boundary negotiation — the boundaries of a phenomenon, the subject as part of relations, what counts as signal versus noise, and the objectivity of data — and offer design starting points for supporting such negotiation.
Apple Intelligence features can be disabled individually on compatible iPhones running iOS 27, even though iOS 27 no longer offers one system-wide switch. Users can turn off Siri AI, summaries, Smart Reply, voicemail suggestions, Writing Tools, image generation and other features through Settings and Screen Time.
What problem it solves
Computer-use projects need environments where agents can operate applications, run code, and interact with graphical interfaces within the same task. They also need repeatable ways to create tasks, evaluate agents, export trajectories, and generate data for training and research.
Key features
Shared desktop Spaces — Cua Spaces gives agents full desktops hosted on computers o…
What problem it solves
Coding agents need a consistent way to inspect, control, debug, and verify applications across mobile, web, TV, and desktop targets. They also need repeatable workflows and reviewable evidence across local devices, simulators, CI, and remote device clouds.
Key features
Inspect and control app interfaces — Agents can read accessibility snapshots, act through refs and selector…
What problem it solves
Hermès Watch Faces require an enabling process on supported Apple Watch devices. Users need instructions for installing or building the app, as well as guidance for preserving data and restoring original settings after updates or removal.
Key features
Install the IPA — The README provides a release-based installation flow using the **Watch** app after sideloading the IPA wit…
What problem it solves
Running large language and vision-language models can require substantial setup and may depend on the available hardware. The project aims to make inference accessible across a wide range of local and cloud hardware with minimal setup.
Key features
Minimal-setup inference — The project provides installation options for common platforms, including scripts, Docker, pre-built b…
What problem it solves
Serving language models requires reading checkpoints, tokenizing requests, and running accelerated kernels while managing a local model server. TensorFold addresses these requirements with a native binary that can run on Apple Silicon and NVIDIA GPUs without requiring Python or MLX for serving.
Key features
Native Metal and CUDA inference — The Zig-based engine executes mode…
What problem it solves
Installing sideloaded apps and importing pairing files can require confusing device setup, certificate handling, and troubleshooting steps. The project provides a desktop workflow for installing IPA files, managing pairing files, and addressing common setup errors.
Key features
App installation — Install SideStore, optionally with LiveContainer, or import any IPA file.
Sourc…
What problem it solves
Sponsor segments, introductions, and advertisements interrupt videos on YouTube for TV clients. A self-hosted application can automate this handling while supporting both network discovery and manual device setup.
Key features
Automatic segment skipping — The application connects to a YouTube for TV app and automatically skips segments such as sponsors and introductions.
Sou…
What problem it solves
Interactive multimodal and speech systems often require generation to span heterogeneous stages with different compute patterns, dependency structures, and resource needs. Serving these pipelines requires coordinated stage lifecycles, inter-stage transport, model integrations, and a unified API surface.
Key features
Coordinated multi-stage runtime — Models generation as coor…
What problem it solves
Users need a command-line workflow to search for apps across Apple platforms and obtain .ipa or macOS .pkg packages from the App Store. The workflow also needs to support App Store authentication, purchases, version metadata, and automation.
Key features
Search across Apple platforms — Search the App Store for iOS, iPadOS, tvOS, watchOS, visionOS, and macOS applications.
Sou…
What problem it solves
Developers need a coding agent that can assist with development work while remaining inspectable and open source. The project addresses the need for configurable agents with different access levels, a straightforward installation path across platforms, and a way to explore codebases without modifying them.
Key features
Two built-in agents with different access levels — A def…
What problem it solves
Developers need to run AI models inside applications while handling model downloads, inference configuration, and chat sessions. The project presents local inference with data privacy and no inference costs as its deployment model.
Key features
High-level application API — The engine exposes a simple API for creating an engine, loading a model, opening a chat session, and re…
What problem it solves
PS5 executables need to be converted into native executables for Linux and Windows, with suitable system libraries available for dynamic linking. The project addresses this porting work without emulation or a separate runtime process.
Key features
Native executable relinking — The relinker converts executables into the native format of the target operating system and provide…
What problem it solves
Time-series forecasting requires models that can forecast univariate and multivariate series across different context lengths. Forecasters also need to incorporate optional covariates and produce quantile forecasts without requiring per-task tuning.
Key features
Univariate and multivariate forecasting — Forecast individual time series or multiple target variates from NumPy a…
What problem it solves
Running AI models locally can require integrating multiple inference engines, model formats, hardware accelerators, and API surfaces. LocalAI addresses this with a small core, on-demand backends, and compatible APIs that work across supported hardware without requiring a GPU.
Key features
Composable inference backends — Backends are packaged separately and pulled only when a…