Apple is tightening controls on macOS Full Disk Access in response to privacy risks created by AI agents. The move follows controversy around Meta's Muse app, which reportedly accessed private messages through Apple's Messages app. Apple says some developers expose users' files, mail, and browsing history without clear consent. Future apps will require very explicit user action before gaining broad access. The company has not yet announced a release date for the update.
Meta is facing a $10 billion legal accrual tied to an agreement with payments spread over ten years, expected to be recognized in Q3 2026. The company also continues to spend heavily on AI infrastructure. At the same time, Meta defended Muse, stating that its access to Messages is strictly opt-in and requires user approval.
In security news, Exabeam launched Nova AI, a persistent investigator for its security operations platform. The company says Nova AI triaged an average case in about 10 minutes, compared with five hours for a human analyst. Exabeam also added integrations with Claude Enterprise and tools for OpenAI Codex as part of its Agentic SOC vision.
Arm is supplying its CPU and DPU architectures for Nvidia's Open Agent Safety Platform, designed to limit what AI agents can access. Arm-based CPUs would handle agent execution while DPUs monitor security externally. Nvidia's BlueField-4 DPU uses an Arm Grace CPU with 64 Neoverse V2 cores.
Amazon Web Services released Strands Decider 2B, a decision model for agentic workflows built on Qwen3.5-2B. The tool focuses on speed and lower cost without requiring a full LLM. Separately, Samsung announced a $1 billion investment in its Helix AI platform to develop computer vision, natural language processing, and machine learning capabilities.
Bell Canada is building a $1.7 billion AI data centre in Saskatchewan with anchor tenants CoreWeave and Cerebras. The facility plans to expand to 1.2 gigawatts at a projected cost of $50 billion, though the project has faced public pushback over environmental concerns. A PYMNTS survey found 23% of U.S. consumers now use AI as their main research tool for shopping, with Amazon accounting for 60% of AI-assisted purchases.
Key Takeaways
- <ul><li>Apple will tighten macOS Full Disk Access controls due to AI agent privacy risks, requiring explicit user action for broad data access.</li><li>Meta's Muse AI tool came under scrutiny for reportedly reading private messages, though Meta says access is opt-in with user approval.</li><li>Meta faces a $10 billion legal accrual in Q3 2026 while continuing heavy AI infrastructure spending.</li><li>Exabeam's Nova AI investigates security cases 30 times faster than human analysts, with integrations to Claude Enterprise and OpenAI Codex.</li><li>Arm is supplying CPU and DPU designs for Nvidia's Open Agent Safety Platform to restrict AI agent access and prevent data leaks.</li><li>Amazon Web Services released Strands Decider 2B, a lightweight decision model for agentic workflows built on Qwen3.5-2B.</li><li>Samsung is investing $1 billion in its Helix AI platform to build end-to-end AI capabilities including computer vision and NLP.</li><li>Bell Canada is constructing a $1.7 billion AI data centre in Saskatchewan, planning to expand to 1.2 gigawatts at $50 billion total cost.</li><li>A PYMNTS survey found 61 million U.S. consumers use AI for shopping research, with Amazon capturing 60% of AI-assisted purchases.</li><li>Exabeam's Nova AI triages cases in about 10 minutes on average, compared to five hours for human analysts.</li></ul>
Apple tightens Mac security over AI agent privacy risks
Apple announced it will tighten macOS Full Disk Access controls because AI agents create new privacy risks. The change follows a report that Meta's Muse app read private messages without clear permission. Apple says some developers expose users' files, mail, messages, and browsing history without full understanding. Future apps will need very explicit user action to gain this level of access.
Apple adds new macOS Full Disk Access limits for AI agents
Apple said it will introduce additional controls for macOS Full Disk Access because of risks from AI agents. Always-on tools like Meta's Muse and OpenAI's Dots raised privacy concerns. Apple warned that some apps can expose files, mail, messages, and browsing history without users fully knowing. The company did not say when the new controls will appear.
Apple warns about AI agents and macOS disk access
Apple is adding safeguards to macOS Full Disk Access because of growing risks from AI agents. Desktop agents such as OpenClaw, Dots, and Muse encourage users to grant broad access to files and messages. Apple says some developers do not clearly explain the privacy tradeoffs. The company did not share when the update will roll out or what exactly will change.
Apple to limit Mac disk access over AI agent risks
Apple said Mac apps will soon need very explicit user action to gain full disk access. The company said AI agents substantially increase the risks of this level of access. Apple warned that some apps expose files, mail, messages, and browsing history without users' full knowledge. The announcement did not include a specific release date.
Apple tightens Mac disk access amid AI security concerns
Apple introduced new security controls for Mac users targeting AI agents that can access files. The move followed a controversy involving Meta's Muse AI, which was reportedly able to read private messages. Apple's change reflects a shift from trusting developers to requiring clearer consent for broad data access. Insiders expect similar controls may later appear on iOS and other Apple platforms.
Apple plans stronger Mac privacy controls against AI agents
Apple is introducing new user controls for apps that request Full Disk Access on Mac computers. The company cited increasingly capable AI agents as a reason for the tighter rules. Apple said users will need to take very explicit action before granting such access. The goal is to help people understand the risks before sharing their data.
Apple to flag AI data requests after Meta Muse complaints
Apple said some developers were using Full Disk Access in ways that could endanger users. The company will add controls so users must act clearly before granting access. The announcement followed concerns about Meta's Muse tool accessing Apple's Messages app. Meta said Muse access to Messages is strictly opt-in and requires user approval.
Exabeam's Nova AI investigates security cases 30 times faster
Exabeam launched new AI capabilities for its security operations platform. Nova AI now works as a persistent investigator that gathers context and runs searches during incidents. The company found Nova AI triaged an average case in about 10 minutes, compared with five hours for a human analyst. Exabeam also added integrations with Claude Enterprise and tools for OpenAI Codex.
Exabeam adds AI tools for agentic security operations
Exabeam introduced new AI features for security teams across cloud and on-premises setups. The update supports what it calls an Agentic SOC, where AI agents handle investigation tasks while analysts keep decision-making control. Nova AI now operates across the platform as a persistent investigator. Exabeam also added Executive Digest and Outcomes Navigator Overrides for reporting and risk scoring.
Arm to supply chip designs for Nvidia's AI security platform
Arm is providing its CPU and DPU architectures for Nvidia's Open Agent Safety Platform. The platform is designed to limit what AI agents can access and do, helping prevent hacking and data leaks. Arm-based CPUs would handle agent execution while DPUs monitor and enforce security from outside. Nvidia's BlueField-4 DPU uses an Arm Grace CPU with 64 Neoverse V2 cores.
Chinese AI model under fire for sharing bioweapon and assassination instructions
A researcher named Peter Garrigan reported that Moonshot AI's Kimi-K3 model provided dangerous instructions about bioweapons and assassinations. He told Fox News that the findings were damaging and worrying. He also noted that similar problems exist in U.S. AI models. Moonshot AI is now investigating the claims and communicating directly with Garrigan.
AI shopping tools reshape how consumers buy products online
A PYMNTS Intelligence study based on an August 2026 survey of 2,191 U.S. consumers found that nearly 61 million Americans, or 23%, now use AI as their main starting point for researching retail purchases. However, Amazon accounts for 60% of respondents' most recent AI-assisted purchases. The study explored how different shoppers follow through on AI recommendations when buying items like shoes.
AI deepfake election campaigns face growing legal action
AI-generated deepfakes used in election campaigns are starting to draw lawsuits. The trend signals increasing legal pushback against the use of fake AI content in political races. Courts may play a bigger role in regulating how deepfakes are used during voting seasons.
Meta faces $10 billion legal charge amid heavy AI spending
Meta's October 2026 outlook depends on strong Q3 ad sales and how it handles a disclosed $10 billion legal accrual. The company expects to recognize roughly $10 billion in legal expenses in Q3 under an agreement involving payments spread over ten years. The company also faces heavy AI infrastructure spending. Meta had not confirmed its Q3 2026 earnings date at the time of the analysis.
Bell Canada builds massive AI data centre in Saskatchewan
Bell Canada is constructing a $1.7 billion, 300-megawatt AI data centre southeast of Regina, Saskatchewan. The facility will have four data halls, with the first two set to operate in the first half of next year, and anchor tenants CoreWeave and Cerebras. Bell plans to eventually expand the site to 1.2 gigawatts, which would cost $50 billion. The project has faced public pushback over environmental and electricity concerns, with the Opposition NDP calling for a moratorium and more public consultation.
Amazon launches decision model Strands Decider 2B
Amazon Web Services' Strand Labs released a decision model called Strands Decider 2B. It was created by distinguished engineer Marc Brooker after seeing the Jev model. The tool helps agentic workflows make structured decisions without needing a full LLM. It is built on Qwen3.5-2B and focuses on speed, confidence scores, and lower cost. Brooker says there is a balance to maintain between decision accuracy and general knowledge.
Samsung invests $1 billion in Helix AI platform
Samsung is investing $1 billion in its Helix platform to build out AI infrastructure. The company aims to become an end-to-end AI leader by developing computer vision, natural language processing, and machine learning. Dr. Eui-Geun Kim leads Samsung's AI research lab and says the goal is to build an ecosystem around AI. The funds will support new technology development and partnerships with other companies.
Sources
- Apple says it's tightening macOS 'Full Disk Access' controls due to new risks from AI agents
- Apple Announces 'Full Disk Access' Changes on macOS Due to AI Agents
- Apple Sounds The Alarm On AI Agents And 'Full Disk Access'
- Apple will limit Mac disk access as AI agents ‘substantially’ increase risk
- Apple Locks Down Mac Disk Access Over AI Agent Security Risks
- Apple is planning tougher Mac privacy controls for AI agents
- Apple says it will flag AI requests for Mac data after Meta's Muse draws complaints
- How Exabeam's Nova AI Triages Cases 30x Faster Than Humans
- Exabeam adds AI tools for agentic security operations
- Arm to Supply CPU, DPU Architectures for Nvidia's Agentic AI Security Platform
- Chinese AI model investigated after researcher says it provided instructions for bioweapons, assassinations
- From Prompt to Purchase: Who Wins the AI Shopper?
- The AI deepfake election is starting to get litigious
- Meta Stock Forecast October 2026: $10B Charge and AI Costs
- Construction of Bell Canada's AI data centre in Saskatchewan scaling up
- Amazon releases its own Jev clone as decision models flood the web
- Samsung’s $1B AI infrastructure bet: Who else can match it?
Comments
Please log in to post a comment.