Business, Deals & Funding
Ars Technica AI
Muse, Meta's extraordinarily privileged AI assistant, has a serious 0-day
Security researcher Patrick Wardle discovered a zero-day vulnerability in Meta's new AI assistant Muse for macOS. The flaw allows any locally installed app or terminal command to access the token authenticating users to their Muse account by exploiting an undocumented setting that changes the transcription endpoint to an attacker-controlled server. Because Muse has extraordinarily broad permissions — including access to WhatsApp, email, calendar, social media, mic, camera, location, and file system — hijacking the agent gives attackers complete control over all of those resources. Wardle demonstrated proof-of-concept attacks including writing malicious files and taking pictures with no user indication. Amazon has also begun blocking Muse from its site. Meta did not respond to questions, despite having published blog posts touting Muse's security design.
Why it matters
This vulnerability illustrates the fundamental tension in deploying highly privileged AI agents: the more useful the agent, the more access it requires, and the more catastrophic a single exploit becomes. Rather than needing to write comprehensive malware, an attacker can simply hijack the agent and leverage its already-granted permissions — a force multiplier that should give the entire industry pause. Storing authentication tokens accessible to any local process and exposing an undocumented s…
Guardian AI
AI is generating a new wave of US populism. It could be a tipping point | Robert Reich
Robert Reich argues that AI is fueling a new wave of economic populism in the US, as evidenced by both Bernie Sanders and Steve Bannon speaking against AI and its corporate backers. He notes that workers' share of the economy has fallen to a record low of 52.9%, while AI promises to further enrich its backers at the expense of jobs and wages. Reich suggests that Trump's alliance with corporate elites has exposed the truth behind corporate America's PR, and that anti-establishment anger directed at wealthy corporate interests—rather than traditional political scapegoats—may represent a political tipping point.
Why it matters
Reich makes a compelling observation about the strange-bedfellows populism emerging around AI, with figures from both the far left and far right converging on corporate criticism. However, the article oversimplifies the economic picture by treating AI primarily as a wealth-transfer mechanism rather than acknowledging its potential for productivity gains that could benefit workers if properly governed. The declining labor share of GDP is a real and serious trend, but it predates AI and has more…
Lenny's Newsletter
🎙️ How I AI: Meta’s Muse review + How Warp ships 2,000 PRs a month with AI factories
This episode of How I AI covers two topics: a hands-on review of Meta's Muse personal AI agent, and how Warp ships 2,000 PRs a month using AI factories. The Muse review highlights its consumer-friendly UX, including frictionless onboarding, smart permission handling, a polished family morning brief generated on the first attempt, ongoing goal tracking (like sleep training), and a transparent activity feed showing all tool calls. Browser-based shopping remains its weak point. The episode also examines Warp's approach to scaling AI-assisted development through systematic 'AI factories' that enable high-volume PR throughput.
Why it matters
The Muse review surfaces genuinely important UX lessons for anyone building consumer AI agents. The permission model — asking at the moment of relevance rather than upfront or constantly — is a design pattern the industry should standardize on. The observation that personal goals require a fundamentally different product model than one-shot tasks is also sharp; most agent products treat everything as a task to complete rather than an ongoing relationship to maintain. The browser-use failures ar…
TechCrunch AI
The man who built Apple’s stores doesn’t buy Silicon Valley’s bet on AI shopping
Ron Johnson, the architect of Apple's retail stores, is skeptical of Silicon Valley's push toward AI-powered shopping (agentic commerce). In an interview with TechCrunch, Johnson argued that while AI will improve the online shopping experience and help consumers become more informed, it won't fundamentally change how people shop — especially for high-consideration purchases like laptops, where buyers want to physically experience the product. He pointed to Apple's retail philosophy, which prioritized in-store human interaction and non-commissioned employees over pure sales pressure, as evidence that physical retail will continue to thrive. Johnson, who is promoting his new book 'Shop Different,' also reflected on his failed stint at J.C. Penney, acknowledging he tried to change too much too quickly. His comments come as Google pushes its Universal Commerce Protocol for AI agents and Ope…
Why it matters
Johnson makes a reasonable point that high-consideration, sensory purchases will resist full AI automation — people genuinely want to hold a laptop or try on clothes before buying. However, his argument feels somewhat self-serving given he's promoting a book about Apple's retail success, and it understates how much shopping has already shifted online for many product categories. The truth likely lies in between: AI agents will probably capture a significant share of routine, commodity purchases…
The Rundown AI
Amazon shuts out Meta's Muse
Amazon has blocked Meta's Muse AI shopping agent just 12 days after its launch, accusing it of browsing Amazon's store without identifying itself and potentially storing customer login credentials — claims Meta denies. The move is part of Amazon's broader effort to wall off outside AI agents that could bypass its lucrative sponsored listings ad business (estimated at $56B), having already taken action against Perplexity, Google, and OpenAI shopping agents. In other news, President Trump announced plans to create an 'AI Force' modeled on the Space Force, with a new AI czar role, to oversee the industry and protect U.S. global leadership in AI while dismissing safety concerns as a hoax — a stance that contrasts with recent calls from Anthropic's Dario Amodei, Sam Altman, and Elon Musk to slow AI development over safety concerns.
Why it matters
The Amazon-Meta clash is the most consequential story here because it signals the real battleground of the AI agent era: platform access and commerce gatekeeping. Amazon's $56B ad business depends on controlling the shopping funnel, and autonomous agents that comparison-shop or bypass sponsored listings pose an existential threat to that model. Amazon's legal and technical blockade is rational self-interest, but if every major platform walls off AI agents, the promise of truly autonomous digita…
The Verge AI
California tightens rules on AI data center energy and water use
California Governor Gavin Newsom signed seven bills that impose new rules on AI data centers to protect residents from bearing the costs of their massive energy and water consumption. The laws require the California Public Utilities Commission to create a new rate classification for data centers, mandate that data centers pay for upgrades to local power grids and water systems, and require proposed facilities to disclose estimated water use and meet energy efficiency standards. Newsom framed the legislation as ensuring that companies profiting from data centers do not do so at the public's expense.
Why it matters
This is a sensible and overdue regulatory move. AI data centers consume enormous amounts of electricity and water, and without rules like these, the costs of infrastructure upgrades get socialized onto local residents who see no direct benefit. Requiring data centers to pay for the grid and water system upgrades they necessitate is a straightforward application of the 'you break it, you buy it' principle. The disclosure and efficiency requirements also give local governments the information the…
Guardian AI
AI could be a bubble and is not yet making Australia more productive, Michele Bullock says
RBA Governor Michele Bullock stated that AI could be a bubble and there is no evidence it is making Australia's economy more efficient. She noted that AI adoption is currently adding to inflation rather than boosting productivity, contradicting the Albanese government's claims that the technology will solve Australia's economic challenges. Her comments came ahead of an expected interest rate rise.
Why it matters
Bullock raises a valid near-term observation: transformative technologies often take years or decades to show up in aggregate productivity statistics, as was the case with electricity and the internet. The 'Solow paradox' — seeing computers everywhere except in the productivity data — is a well-documented phenomenon. However, calling AI a potential bubble conflates two distinct questions: whether AI companies are overvalued in markets, and whether the underlying technology will deliver long-ter…
TechCrunch AI
OpenAI forms math advisory group as its AI resolves more than 100 open problems
OpenAI has announced a new independent Advisory Group on Mathematics and Artificial Intelligence, hosted at the Institute for Advanced Study in Princeton. The group of nine prominent mathematicians will assess the significance of AI-generated mathematical results and coordinate their release. This follows OpenAI's claimed resolution of over 100 open math problems, including the Navier-Stokes Millennium Prize problem, which has drawn criticism from 25 Fields Medal winners who signed an open letter arguing AI labs are threatening mathematicians' intellectual work. Notably, the advisory group will have no power to slow down or redirect OpenAI's math research, serving only in an advisory capacity without decision-making authority over any AI company.
Why it matters
This advisory group appears to be a carefully constructed PR exercise rather than a meaningful governance mechanism. The most telling detail is buried in the announcement itself: the group explicitly cannot influence the pace or direction of OpenAI's mathematical research. An advisory body that cannot advise on the most consequential question — whether and how fast to proceed — is advisory in name only. It exists to provide a veneer of community engagement while OpenAI continues racing to publi…
The Verge AI
Can John Ternus find Apple’s next big thing?
The Verge's Nilay Patel interviews Bloomberg's Mark Gurman about Apple's leadership transition from Tim Cook to new CEO John Ternus. The discussion centers on Ternus's first major product launch — the iPhone Duo, Apple's first foldable smartphone — and what it signals about Apple's strategic direction. They explore how Cook orchestrated the leadership transition over the past year, potential changes to Apple's culture under Ternus, and the broader challenge of navigating Apple from the iPhone era into the AI era, including the state of Siri and Apple's AI ambitions.
Why it matters
The timing of Apple's CEO transition alongside the launch of its first foldable device is a shrewd piece of corporate storytelling — it gives Ternus an immediate, tangible 'new era' narrative rather than inheriting a continuity product cycle. However, the real test won't be hardware novelty like foldables, which competitors have shipped for years. It will be whether Ternus can fix Apple's genuinely lagging AI and Siri capabilities, where the company has fallen meaningfully behind. A hardware-fo…
Guardian AI
Halt super-intelligent AI with non-proliferation treaty, Ed Davey to say
Ed Davey, leader of the Liberal Democrats, is calling for a global nuclear-style non-proliferation treaty to halt the development of super-intelligent AI. He plans to make this proposal in his keynote speech at the Lib Dem conference in Brighton, while accusing Andy Burnham of relying on Donald Trump and 'tech bros' to address AI-related challenges.
Why it matters
The comparison to nuclear non-proliferation is thought-provoking but faces significant practical challenges. Nuclear weapons require rare materials and massive infrastructure, making proliferation trackable and containable. AI development, by contrast, is fundamentally software-driven and increasingly democratized — the 'materials' are compute, data, and algorithms, all of which are widely distributed. A treaty modeled on nuclear non-proliferation would struggle with verification, enforcement,…
TechCrunch AI
Discover what’s next: 5 days left to save up to $200 on your TechCrunch Disrupt 2026 ticket
TechCrunch is promoting its Disrupt 2026 conference (October 13-15, Moscone West, San Francisco) with a limited-time discount of up to $200 off tickets, plus 50% off a second ticket, ending September 25. The article highlights the 'discovery' value of attending: seeing 200 early-stage startups selected by TechCrunch (Startup Battlefield 200), watching 20 of them pitch live for the Disrupt Cup and $100,000 in equity-free funding, and networking with 10,000+ founders, investors, and operators.
Why it matters
This is straightforward promotional content for TechCrunch's flagship event, dressed up as editorial. The 'reason to attend' format is a common drip-marketing tactic to create urgency around a deadline. That said, Disrupt has historically been a legitimate venue for startup discovery — companies like Dropbox and Mint famously launched there. The real question is whether the event justifies its price tag in an era where startup demos, investor panels, and networking can happen online. For people…
The Verge AI
iPhone owners can now submit claims in Apple’s $250 million Siri AI settlement
Apple is paying $250 million to settle a class action lawsuit alleging it misled consumers about AI-powered Siri features for the iPhone 16 lineup. US residents who purchased an iPhone 15 Pro, iPhone 15 Pro Max, or any iPhone 16 model between June 10, 2024, and March 29, 2025, can now file claims on the settlement website. Eligible claimants may receive an estimated $25 per device, potentially increasing to $95 depending on participation rates. Claims require a name, contact information, and iPhone serial number, with a December 21, 2026 deadline. The lawsuit argued Apple created reasonable consumer expectations that AI features previewed at WWDC 2024 would ship with the iPhone 16, but the device launched with limited capabilities despite being marketed as 'built for Apple Intelligence.' The AI-upgraded Siri only arrived with iOS 27 in September 2026. Apple has denied wrongdoing.
Why it matters
This settlement highlights a growing accountability gap in tech marketing, where companies announce ambitious AI features to drive hardware sales before those features are actually ready. Apple marketed the iPhone 16 as 'built for Apple Intelligence' when key capabilities were months — ultimately over two years — away from shipping. A $250 million settlement sounds significant, but at $25-$95 per device across millions of eligible phones, it's a modest cost for Apple relative to the revenue tho…
From X/Twitter
- Cloudflare open-sourced its internal Security Audit Skill for AI agents — a six-phase pipeline where the agent that finds a vulnerability never validates it, cutting false positives.
- Dickie Bush's 8-step guide walks through building and shipping a landing page with Codex and Vercel, from first-draft copy to custom domain.
- GPT Image 2.5 on Wavespeed turns a selfie into a miniature world with a single prompt.
- Warp CEO Zach Lloyd gets specific on how to actually run an AI software factory: ticket intake, human-touch-per-PR metrics, evals for failure modes, and self-improving harnesses.
- Codos bills itself as the first virtual Chief AI Officer — interviews employees, deploys cross-functional automations, and runs on your own servers.
- Noah Kagan says AppSumo nearly fully automated its $57k/month Google Ads operation using Claude Code, n8n, and the Meta and Google APIs — replacing both a Google and Facebook agency.
- Matt Cassinelli, after 10 years in Shortcuts: headless automation still fails without dedicated UI — context collapse is real, and nothing has solved it.
- Colin McDermott built a Grok Bot × Jev classifier template to give agents fast, calibrated signal out of the box.
- Tom Alder launched a smart inbox that unifies DMs from LinkedIn, Slack, WhatsApp, X, and iMessage, stores everything locally, and drafts AI replies in your voice.
- Jensen Huang: AI companies lobbying for new regulation are really asking to be relieved of existing laws, and that's a problem.
- Background computer use in Claude Code's Mac desktop is now in beta — Claude sees and acts in approved apps while you keep working, on Pro and Max plans.
- Grok 4.7 ships as a notable step up from 4.6 at the same price and speed.
- Apple fixed "Describe a Shortcut" in macOS 27.2 beta to support actions from third-party apps, not just Apple's own.
From Reddit/HN/YC
- [Hacker News] TypeSafe AI's Jev goes head-to-head with GPT-6 Astra in Vercel's structured output and latency benchmarks.
- [Hacker News] Rest of World argues the US is framing the AI race with China around the wrong metrics — trust gaps matter more than benchmark gaps.
- [Hacker News] Google is extending Gemini 2.5 access indefinitely for active API users, sidestepping the usual deprecation timeline.
- [Hacker News] Icon Solutions makes the case that AI apathy in financial services is now a larger risk than AI hype.
- [Hacker News] Michael Lynch's guide to making your code reviewer's job easy — smaller PRs, clear context, no leaving detective work to the reader.
- [Hacker News] Tom's Hardware finds the Mac Studio M5 Ultra outpaces both DGX Spark and Threadripper for local model inference.
- [Hacker News] Tesla recorded four fatal Autopilot and FSD crashes in July 2026 — the deadliest single month on record.
- [Hacker News] Scientific American raises doubts about whether OpenAI solved the right Navier-Stokes problem and what the headline result actually proved.
- [Hacker News] C3 0.8.4 is the last keyword renaming before the syntax freeze — the language is locking in its final form.
- [Hacker News] Research shows how progressively revealing code shapes programmers' visual attention — with implications for AI-assisted editor design.