AI & SaaS News, Nov 4: ChatGPT 1.2B Users, Siri AI Opens

Last checked November 4, 2026 · News roundup

Today in brief: OpenAI says ChatGPT now reaches 1.2 billion people a week, up 200 million in about two months, and Apple has opened Siri AI to every developer through App Intents in the iOS 27.3 beta. Anthropic is extending its invisible text watermark to four older Claude models and published FlashProve, an open-source way to prove which model wrote a reply, while a GitHub report says the Postman desktop app quietly installed plugins into Claude Code and Codex. In SaaS and business news, CrowdStrike is buying two AI security startups for about $495 million, Reddit is testing ads in search results, and inference host Fal doubled its valuation to $16 billion.

Part 1 · AI models, apps and features

How many people use ChatGPT? OpenAI says 1.2 billion a week

ChatGPT went from 1 billion to 1.2 billion weekly active users in 60 days, according to Adam Fry, who leads ChatGPT's consumer product. He shared the figures at an OpenAI dinner in New York on Monday, November 2, as reported by Alex Kantrowitz's Big Technology newsletter on November 3.

Fry gave three more numbers. People create about 1.5 billion images a week, voice mode and dictation together have 250 million weekly users, and search-style questions in ChatGPT now equal roughly one-fifth of Google's search volume. That last figure comes from OpenAI's internal analysis and has not been checked independently.

He also said OpenAI plans to merge ChatGPT's separate Chat and Work modes into one interface "soon", without giving a date. Nothing changes for paying users today, and no plan prices moved.

Why it matters: for marketers, this is the clearest sign yet that AI answers are a search channel of their own. If your product or content is not showing up when people ask ChatGPT for recommendations, that gap is now worth measuring alongside Google rankings.

Siri AI for developers: any app can plug in with the iOS 27.3 beta

Apple opened Siri AI to all developers on November 3. With a new developer preview in the iOS, iPadOS, macOS, visionOS and watchOS 27.3 betas, apps can expose their content and actions through App Intents, so users can ask Siri in plain language to check a delivery, book a table or place an order. A new "skills" option lets developers chain several actions into one request, and Siri AI can now reach any item in the Spotlight semantic index, not only items that fit Apple's preset schemas.

Apple's own apps go first: Calendar, Files, Finder, Journal, Mail, Maps, Messages, Music, Notes, Photos, Preview, Reminders, Safari, Shortcuts and Weather. MacRumors notes that in the beta Siri can unsubscribe you from mailing lists in Mail and close duplicate tabs in Safari.

Supported languages are English, French, Japanese, Korean, Portuguese and Spanish, with Traditional Chinese, Norwegian, Turkish and Vietnamese coming. Siri AI is not available in the EU on iOS, iPadOS or watchOS for now, and server-side requests carry daily limits, with paid expanded access promised later.

What to do: if you ship an iPhone or Mac app, install the 27.3 beta and map your most common user tasks to App Intents now, because nothing can go to the App Store before spring.

Does Claude watermark its text? Four older models were due to join on November 3

Yes, for most current models, and the list is getting longer. Anthropic already adds an invisible statistical watermark to text from its newer models, and an Anthropic notice to users reported by Mashable said four older ones would join on Tuesday, November 3: Claude Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6. The watermark applies worldwide with no opt-out, adds no visible characters or extra tokens, and Anthropic says it does not change answer quality.

The change is for the EU AI Act, whose Article 50 requires machine-readable marking of AI output. Models launched after August 2, 2026 had to support marking at launch, and older ones get a transition period. When we checked Anthropic's help center on November 4, its table still showed the four older models without text watermarks, so the page may lag the rollout.

Ordinary users and API customers cannot detect the mark. Detection is a private preview for regulators, researchers, media, fact-checkers and companies with compliance duties. A detected mark only suggests Claude was involved at some point; heavy rewriting removes it, and short or code-heavy text carries little signal.

Did Postman change my Claude Code and Codex settings? What the GitHub report shows

Possibly, if you run the Postman desktop app on the same machine as an AI coding agent. A GitHub issue opened on October 25 against Postman desktop 12.33.4, which drew wide attention on Hacker News this week, says the app installed a Codex plugin and a Claude Code plugin without asking. The reporter's logs label the install "auto-existing-upgrade".

According to the report, the installer added a Postman marketplace and enabled its plugin in the Codex config, enabled a Postman plugin in Claude Code's settings, put a Postman CLI in the user's local bin folder, and registered a startup hook that tells the agent to load a Postman skill before API work. The reporter says the installer also wrote a trust record for its own hook, though they note this does not prove who created it. As of November 4 the issue had no visible reply from Postman staff and no fix version.

What to do: look for these entries, remove anything you did not approve, and review agent hooks the way you would review a browser extension.

What is FlashProve? Anthropic's open-source proof of which model answered

FlashProve is a research prototype that lets a user check, with cryptography, that a reply really came from the model a provider claims to be serving. Anthropic published it on November 2 with researchers from EPFL, MIT, StarkWare, Succinct and RESI. The provider first publishes a commitment, a kind of fingerprint, of the model weights; each reply then comes with a zero-knowledge proof that can be checked against that fingerprint without revealing the weights.

It is not cheap yet. On a 400-billion-parameter mixture-of-experts model, proving ran at about 110 tokens per second on one H100 GPU, each 1,024-token request produced a proof of about 19 MB, and proving cost roughly 70 to 95 times the GPU time of normal inference. Checking a proof took about 17 seconds on a 24-thread CPU. The code is on GitHub under the Apache 2.0 license.

No lab, Anthropic included, has plans to deploy it yet, TechCrunch reports. The idea matters for anyone who pays for a premium model through a reseller or router and wants proof they are not getting a cheaper substitute.

Netflix AI Discover: the mood-based tab rolling out on US TVs

Netflix launched an "AI Discover" tab on November 3 that suggests titles based on your mood, the time you have, or what is trending. You pick a theme such as "something light", "scare me", "family movie night" or "quick watches", narrow it down, or hit "something different" to start over. A new "Tune Your Recommendations" button next to Play and My List lets you save a title or hide it from future picks.

It is rolling out in the US to people who use Netflix in English on supported TVs, and Netflix says more users get it in the coming weeks. Look for a "Discover" button with a red gem icon and a "New" label in the top menu. Netflix calls the tab a "test kitchen", so features that do well may move into the main app. No plan change or extra fee was announced.

Part 2 · SaaS & business

CrowdStrike buys Fiddler and Dreadnode: about $495 million for AI agent security

CrowdStrike is buying two startups that secure AI agents, TechCrunch reported on November 3, citing The Wall Street Journal for the prices: about $400 million for Fiddler AI and about $95 million for Dreadnode. SecurityWeek reports the deals are expected to close by January 31, 2027.

Fiddler, founded in 2018, makes software that monitors how AI agents behave inside a company and will become Falcon Observe on CrowdStrike's platform. Dreadnode, founded in 2023, automates attack testing of AI models, apps and agents before they go live; its tools will feed into Falcon Guardian and its security policy recommendations.

Who it affects: current Fiddler customers should expect the product to move under CrowdStrike branding and contracts over the coming months, so ask your account team about roadmap and pricing before renewal. For everyone else, it confirms that "agent security" is becoming a standard line item next to endpoint and cloud security.

Reddit search ads: a new test after a drop in US users

Reddit started testing ads on its search results pages on November 3 with "a select group of eligible advertisers", TechCrunch reports. A search for "home office setup" could show ads for desks, monitors or chairs, matched to the query.

The test follows a weak forecast. Reddit guided fourth-quarter revenue to $1.00 billion to $1.03 billion, below the roughly $1.05 billion analysts expected. Global daily users were flat at 130.3 million, while US daily users fell from 53.2 million in the second quarter to 50.7 million, and Reddit warned the US figure could drop by a low double-digit percentage by the fourth quarter. The company is also moving more content behind a login, starting outside the US, which ties into the access changes we covered in our note on the Reddit RSS feed shutdown.

For advertisers: search ads reach people at the moment they look for a product, which is closer to Google Ads intent than to Reddit's usual feed ads. If you sell to Reddit's audience, ask your rep about joining the test.

Fal raises $1.3 billion at a $16 billion valuation, and Traversal reaches $1 billion

Fal, the San Francisco company that runs image, video and audio models for developers, raised a $1.3 billion Series F led by Bessemer Venture Partners, with Nvidia, Sound Ventures and T. Rowe Price joining, TechCrunch reported on November 3. Bloomberg puts the valuation at $16 billion, double April's $8 billion, and says Fal passed $800 million in annualized revenue this year.

Also on November 3, Traversal raised a $120 million Series B at a $1 billion valuation, co-led by Kleiner Perkins, Redpoint and Khosla Ventures, bringing total funding to $173 million. Its AI site reliability software finds the root cause of production incidents; Nubank, Harvey, DigitalOcean and PepsiCo are named customers.

Why it matters: money is flowing to the plumbing under AI apps. If you build on hosted open models, competition among inference providers such as Fal should keep pushing per-request prices down.

Saazora take: three checks before the weekend

First, if your team uses coding agents, audit plugins and hooks after installing any desktop tool; the same habits in our checklist for vetting Claude skills apply to plugins added by other apps. Second, if you build apps, the Siri AI preview and ChatGPT's 1.2 billion weekly users point the same way: people increasingly reach products through an assistant, so design your key actions so an AI can call them. Our list of the best AI app builders, part of the AI app builders hub, compares tools that can get a prototype out quickly. Third, if you publish AI-assisted text for EU readers, assume Claude output from most models now carries a watermark, and keep your own disclosure policy current.

People also ask about today's news

Can I tell if a text was written by Claude?

Not on your own. Anthropic's text watermark can only be read through a detection API that is in private preview for regulators, researchers, media and companies with compliance duties. Its free checker reads Content Credentials on files, not text. Even a positive result only shows Claude was likely involved, not that it wrote everything.

When can I use the new Siri AI on my iPhone?

Apple says Siri AI ships to customers "in a few months". Right now it is in the iOS 27.3 developer beta. You need an Apple Intelligence device set to a supported language, and it is not offered in the EU on iPhone for now. App Store apps can add Siri AI features from spring.

Is Postman safe to keep installed?

Postman has not responded publicly to the GitHub report as of November 4. If you also use Claude Code or Codex, check your agent config files for Postman plugins and hooks you did not approve, and remove them. Keeping Postman is your call, but review its integrations after each update until the company explains the behavior.

How do I get Netflix AI Discover?

It is rolling out gradually to US members who use Netflix in English on supported TVs. If you have it, a "Discover" button with a red gem icon and a "New" label appears in the top menu. There is no setting to switch it on, and Netflix says more members get it in the coming weeks.

Sources: Big Technology, Nov 3, 2026 · Apple Developer News, Nov 3, 2026 · MacRumors, Nov 3, 2026 · Anthropic help center, checked Nov 4, 2026 · Mad Robot (citing Mashable), Nov 1, 2026 · GitHub issue 14370, opened Oct 25, 2026 · Anthropic research, Nov 2, 2026 · TechCrunch (FlashProve), Nov 3, 2026 · TechCrunch (Netflix), Nov 3, 2026 · TechCrunch (CrowdStrike), Nov 3, 2026 · SecurityWeek, Nov 3, 2026 · TechCrunch (Reddit), Nov 3, 2026 · TechCrunch (Fal), Nov 3, 2026 · Unite.AI (Traversal), Nov 3, 2026