Get tomorrow's brief in your inbox
Today: Zapier added Muse Spark reasoning to 9,000+ workflow integrations, Apple locked down Mac disk access after AI agents started reading private messages, and Meta open-sourced Muse gadget firmware so you can put AI on your TV or Raspberry Pi.
Apple rewrites Mac security after AI agents overstepped
Apple announced it's tightening macOS "Full Disk Access" controls, a sweeping permission originally designed for backup apps that's now being exploited by AI agents. The move comes after Inc. columnist Jason Aten reported that Meta's Muse AI somehow knew the contents of his private messages despite him not giving explicit permission. Apple's response is blunt: "Some developers are using Full Disk Access in ways that could put users at risk, exposing everything on their systems - including files, mail, messages, and even browsing history - without users' full knowledge and understanding." The company added that as AI agents become "increasingly capable and autonomous, the risks associated with this level of access will grow substantially."
For small businesses running AI tools on Mac, this matters. If you're using desktop AI agents (Muse, Dots, or similar), check your System Preferences > Privacy & Security > Full Disk Access settings. Apple says future versions of macOS will require "very explicit user action" before granting this level of access. Until then, treat any AI agent requesting full disk access like you'd treat handing over your office keys: assume it can read everything, including sensitive client data, financial records, and internal communications. Source: TechCrunch, Ars Technica
Zapier brings Muse Spark reasoning to existing workflows
Zapier's AI by Zapier now supports Meta's Muse Spark 1.3 model, letting you drop AI reasoning steps into workflows you already run. This is different from using Muse directly: instead of chatting with an agent, you embed Muse's reasoning into specific automation steps where you actually need judgment - like summarizing support tickets, scoring inbound leads, or drafting follow-up emails based on call transcripts. Cost comes out of your Zapier task count, not a separate Meta bill. If you're already using Zapier to connect your CRM, project management tool, and email, this gives you AI reasoning without rebuilding your stack around a new chatbot. Source: Zapier
Gemini Enterprise now connects to 100+ business apps
Google added over 100 Gemini Enterprise connectors, covering Confluence, Jira, Salesforce, ServiceNow, SharePoint, and Zapier (which extends reach to 9,000+ apps). This is for Google Workspace customers on Gemini Enterprise plans - not the free Gemini app. Admins set up connectors in the Gemini Enterprise console; once enabled, employees can ask Gemini to pull data from those sources while respecting underlying permissions. Practical use case: "Check my Gmail for the customer's latest message, draft a reply, and add a follow-up task to Jira." If you're evaluating whether Gemini Enterprise is worth it for your team, check the connectors directory to see if your tools are covered. Source: Zapier
Meta open-sources Muse Gadgets (and ships 5,000 free ones)
Meta released open-source firmware and a Linux SDK for building custom Muse-powered hardware. Projects include loading Muse onto a color e-ink display, an HDMI stick for your TV, or a Raspberry Pi. Meta also built 5,000 "Muse Home Link" USB-C devices that let Muse control smart home gear, and is giving them away free to Muse subscribers (while supplies last). This is a niche play, but it signals Meta's strategy: make Muse the default AI layer across every device and surface you own, not just your phone. If you're a tinkerer or want AI embedded in custom hardware for your business, this is a free playground. Source: TechCrunch
OpenAI's Dots: Workplace AI that can also order burritos
OpenAI launched Dots, an AI agent platform that runs on a virtual machine and can access apps like Blender, GIMP, and your own computer (via the desktop ChatGPT app). Unlike Muse's personal shopper vibe, Dots feel like enterprise software: you get one Dot, it has a blobby avatar, and it's designed for work tasks (though it can also handle personal errands). The catch: Dots are available first on OpenAI's $100/month Pro plan, and early testing shows mixed results - one reviewer got stuck at a "human check" when trying to schedule an internet install, and Dots couldn't find a free trial option that a competitor agent found instantly. If you're already paying for ChatGPT Pro for your team, Dots might be worth testing for repetitive tasks. Otherwise, wait for the kinks to get ironed out. Source: The Verge
Google shipped Gemini 4 Argon, 3.8 Flash, and expressive TTS
Google's September AI roundup: Gemini 4 Argon is their newest frontier model built for cybersecurity defense and complex reasoning, with a 1-million-token output limit. It's currently rolling out to trusted cyber defenders through the Fairwind Program, not broadly available yet. Gemini 3.8 Flash and 3.8 Flash Cyber bring reasoning and coding performance to agentic workflows and autonomous vulnerability discovery. Gemini 3.8 Live and 3.8 Flash-Lite TTS let you generate custom voices and direct scene dialogue from text prompts. If you're building voice-first applications or need AI for security work, these models are worth tracking. Pricing and broad availability vary by model. Source: Google AI Blog
OpenAI released GPT-6 model guide
OpenAI published a practical guide for GPT-6 models, covering how to choose a model (Astra for hardest reasoning, Sol for complex coding, Luna for focused tasks at scale), set reasoning levels (Low/Medium/High/Max), and manage context with caching and compaction. Key takeaway: treat reasoning level like an intelligence/price tradeoff. Use Low for routine tasks, Medium for work requiring judgment, High for difficult debugging, and Max only when High falls short and the improvement justifies the cost. If you're building on GPT-6, this guide will save you money and latency by matching the right model and reasoning level to each task. Source: OpenAI
Chatham Financial cuts trade validation from 30 minutes to under 4 with Codex and GPT-5.6
Chatham Financial, which advises clients on capital markets decisions, used OpenAI's Codex to build a trade validation app that gathers transaction evidence, compares key terms, and flags discrepancies for review. Early measurement shows review time dropped from approximately 30 minutes to under 4 minutes. Chatham is comparing the app's results with those of experienced reviewers before expanding automation. The firm also uses GPT-5.6 to power AI features in its Chatham Vibes platform, where employees build custom apps for client-facing workflows like reviewing maturing-cap trades and preparing pricing workbooks. Business impact: faster, more auditable results with professional oversight still in place. If you're in financial services or any industry where accuracy and auditability are critical, this is a model for how to adopt AI without cutting humans out of the approval loop. Source: OpenAI
GTM engineers are the new role AI created
Clay coined the term "GTM engineer" in 2023 to describe practitioners who build automated revenue workflows instead of executing individual sales and marketing tasks by hand. The role is spreading: roughly 100 GTM engineering job listings now appear each month, including at companies like Cursor, Lovable, and Webflow. What GTM engineers do: use AI, data, and automation to research prospects, qualify leads, personalize outreach, and move information between systems - work that once required multiple people and tools. If you're deciding whether to hire more people to execute your existing sales process or build systems that let your current team execute differently, this session at TechCrunch Disrupt (October 13-15) will give you a framework. Clay itself tripled ARR to $100 million and announced a $115 million Series D at a $7.1 billion valuation. Source: TechCrunch
AI hallucinations are making entitled customers worse (and dangerous)
A New York City server reported that diners are using ChatGPT to verify allergen warnings, then ignoring her when she tells them a dish contains shellfish. "They just want to talk to ChatGPT," she said. "You're allergic to shellfish, and ChatGPT says there's no shellfish in this, and we are telling you, yes there is." Other workers report customers requesting products that don't exist, AI-generated itineraries that schedule river rafting upstream (not possible), and parents turning to ChatGPT for childhood development advice. Business lesson: if you run a service business, train your team to handle customers who trust AI over your expertise. Have a clear, calm script for when AI hallucinates: "We have this really specific menu, maybe it's showing you a dish from a different restaurant." Don't argue with the chatbot; redirect to facts you can verify. Source: The Verge
White House rebranded AI as "super intelligence" (seriously)
President Trump signed an executive order officially rebranding AI as "super intelligence" and got nearly every major tech CEO (Zuckerberg, Bezos, Musk, Anthropic's Dario Amodei) to sign an AI safety pledge Trump called "morally binding." Despite the rebrand and the pledge, the biggest money in AI still appears to be coming from enterprise, not consumer. According to TechCrunch's Equity podcast, only 2% of consumers are buying into consumer AI products. For small businesses, this means: don't chase consumer AI hype. Focus on enterprise tools where the ROI is clearer. Source: TechCrunch
Data center backlash is spreading from the U.S. to Europe and Asia
Public opposition has impacted around $42 billion of data center investments in Europe (delays and cancellations) and is growing in South Korea, where residents are protesting data centers near residential areas. Amazon AWS CEO Matt Garman published a 3,000-word blog warning communities not to block data centers, saying "our nation can't afford to lose" the AI race. He added: "Right now there are over 100 data center moratoriums being considered across the country. If these measures are enacted, the US could be writing its own losing ticket to this race." Microsoft is responding with "biomimicry" - making data centers greener with native trees and gardens - but residents want more than landscaping. If your business depends on cloud infrastructure, watch for power constraints and delays in regions facing backlash. Source: CNBC, The Verge
Stability AI pivots to music (with major label backing)
Sean Parker (Napster co-founder) and CEO Prem Akkaraju are rebuilding Stability AI around music, backed by $76 million from Sony, Warner, and Universal (who also licensed their catalogs for training). Stability released three new audio models and AI music-editing software; an upcoming update will let users hum a melody or beatbox a drum pattern to steer the AI. Parker says he's "playing by the rules this time" after Napster. If you're in music production or content creation, this is a licensed, legal AI music tool with major label backing - worth watching as it matures. Source: TechCrunch
Use AI to turn blog posts into social copy
Feed a new article to Muse Spark (via Zapier's AI by Zapier) and get platform-ready drafts for LinkedIn and X, then send them straight to your social scheduler. This takes a multi-step manual process (read article, extract key points, rewrite for each platform, schedule) and collapses it into one automated workflow. Set it up once, and every time you publish a blog post, you get social copy without lifting a finger.
Triage support tickets with Muse Spark
Have Muse Spark read incoming tickets, assign a priority level, and route anything urgent to the right Slack channel before a human opens it. This gives your team a head start on urgent issues and keeps low-priority tickets from clogging the queue. Use AI by Zapier to drop Muse Spark reasoning into your existing support workflow.
This week revealed the gap between AI's promise and its readiness for unsupervised use. Apple locked down Mac permissions after agents read private data, customers are trusting hallucinated allergen warnings over human expertise, and OpenAI's Dots got stuck at a "human check" during a simple scheduling task. At the same time, Zapier and Google are making AI accessible across thousands of business apps, and companies like Chatham Financial are seeing real ROI by keeping humans in the loop for judgment and approval. The pattern: AI works when it's embedded in governed workflows with human oversight, not when it's given free rein. Watch for more security and privacy guardrails as desktop AI agents mature, and for the gap between enterprise adoption (where ROI is clear) and consumer adoption (still at 2%) to widen.