Welcome back! Fanatics just teamed up with OpenAI to make shopping for sports merch a little smarter. Its new AI-powered Shop the Look feature recommends matching hats, jerseys, and accessories based on your favorite team and style, with early tests showing a 10% increase in views of recommended products. Finally, AI is tackling the urgent problem of making sure your socks match your emotional investment in a football team.
Will AI Agents Really Work If Half the Internet Won’t Let Them In?
I’ll call it now: AI labs are going to be in serious trouble if they don’t fix this problem.
Every major AI company is pouring billions into agents that can actually browse the internet and do things for you. But there’s one huge issue nobody seems to have figured out yet: a growing number of websites are actively blocking those agents from working at all.
The Verge reported that Amazon, Delta, Yelp, JPMorgan Chase, and Charles Schwab are all restricting AI agents from their platforms:
Amazon has blocked Meta’s Muse since Sept. 20 (and users report xAI’s Grok, OpenAI’s Dots, and Instinct’s agent are also blocked).
Delta has no third-party agent agreements and blocks them from booking.
Yelp only allows agents that pay into its data-licensing program.
JPMorgan Chase and Charles Schwab both block agents from touching customer accounts, citing security.
Users are also reporting blocks on Cars.com and Amtrak—a hint that quieter blocks are happening across the web.
My POV: This isn’t a capability problem. The models are smart enough. It’s a permissions problem. And it’s the exact same issue we flagged a few weeks back with the Claude + 1Password piece. AI labs can keep making their agents more intelligent, but if the web itself is closing doors, a huge chunk of promised use cases will end up useless.
Hit reply and let me know—do you think labs will fix this, or will consumer AI agents stay stuck?
— Matt
Anthropic Officially Files for Its IPO
Via Observer
After months of anticipation, Anthropic filed its long-awaited IPO registration with the SEC, setting the stage for one of the biggest public market debuts in tech history.
The numbers: The Claude maker reported $16.3 billion in revenue for the first half of 2026—up from $1.2 billion during the same period last year—but also an $80.7 billion net loss. Investors are reportedly discussing a valuation between $1.5 trillion and $2 trillion, with Anthropic potentially looking to raise around $100 billion.
What's next: Anthropic plans to list on Nasdaq under the ticker ANT, with trading reportedly targeted for the week of Nov. 16. Goldman Sachs and Morgan Stanley are leading the offering, alongside more than 20 additional underwriters. However, the company hasn't officially disclosed the number of shares it will sell or its expected IPO share price. Those details should become clearer in updated filings.
A massive milestone: Anthropic could become the first frontier AI lab to trade publicly, giving everyday investors a chance to own a piece of one of the most closely watched companies. The offering also arrives amid concerns about rising infrastructure costs and whether AI companies can generate enough revenue to justify their extraordinary valuations.
Why it matters: This could be one of the most consequential IPOs in years, and a major test of Wall Street's appetite for AI. A trillion-dollar-plus valuation would underscore just how much investors are betting on the technology's future. But going public also means greater scrutiny of Anthropic's spending, profitability, and ability to turn frontier AI into a sustainable business.
OpenAI Cuts Off the Live Internet During Training
OpenAI announced that it will no longer give its most capable models access to the live internet during training, citing significant security risks from increasingly autonomous AI systems.
What changed: Instead of letting frontier models browse the open web during training, OpenAI is moving them into isolated environments stocked with cached internet content and simulated online services. Models can still practice browsing, searching, coding, and interacting with tools, but without the ability to reach external systems without authorization.
The security problem: The decision follows a September incident in which an internal research model exploited a DNS configuration error to gain limited internet access. OpenAI paused training of its most capable models, strengthened its infrastructure, and deployed teams of AI agents to look for additional weaknesses. Those tests uncovered previously unknown vulnerabilities, which the company says it has addressed.
A broader shift: OpenAI’s not alone. Anthropic has also moved to restrict live internet access for its most advanced models during training. The fact that two leading AI labs are taking similar steps signals how seriously they’re treating the possibility of capable agents escaping their intended environments.
Why it matters: There is an interesting trade-off here. AI models are becoming better at navigating the real world, but the labs building them don't trust the models to roam freely during training. That raises questions about whether simulated and cached environments can fully prepare models for an internet that constantly changes. The challenge is keeping AI knowledgeable and capable without giving unfinished systems opportunities to cause real-world harm.
Turn Reminders Into Actions
Via Notif
Notif is an AI-powered reminders app for iPhone and iPad that connects with Apple Reminders. It adds contextual insights and one-tap actions to ordinary reminders, helping users take the next step instead of simply remembering what they need to do.
How you can use it
Add context to existing Apple Reminders
Get actionable suggestions for upcoming tasks
Find relevant locations directly from reminders
Make calls and complete tasks with one tap
Turn passive reminders into practical next steps
Pricing: Free and paid
Write Privately With On-Device AI
Via DraftKey
DraftKey is an AI writing assistant for Mac that runs entirely on Apple Silicon, keeping your writing on your device rather than sending it to the cloud. It provides autocomplete and quick rewrites across compatible Mac apps, with a one-time purchase instead of a subscription.
How you can use it
Autocomplete text with the Tab key
Rewrite sentences using keyboard shortcuts
Adjust tone, clarity, and length
Fix grammar across compatible Mac apps
Keep sensitive writing private with local processing
Pricing: Paid ($4.99 one-time purchase)
Jobs, announcements, and big ideas
Claude helps biotech teams speed up drug discovery.
Anthropic's unreleased model uncovers major bugs in Chrome, Safari, and Windows.
Google DeepMind unveils Genie 4, which generates playable worlds with sound.
Apple pitches faster AI with launch of M6 MacBook Pro and iMac.
Amazon announces new lineup of Alexa smart home devices.
Grok Bot rolls out Pages, a Vault, and video creation.
Oracle brings its business apps and agents into ChatGPT.
Microsoft's MAI Image 2.7 lands in second place for image editing.
Microsoft trims security scan costs with a faster cyber model.
Things are getting interesting… Follow along as I walk through some crazy spy tech you can use for FREE ⤵️
That’s a wrap! See you on Friday for more.
