Siri AI opens to apps as Google keeps school AI off

Apple opened Siri AI to app developers on Tuesday, Nov. 3, in a preview for testing, and said it will reach customers in a few months. The same day, Google said AI tools in K-12 student accounts will stay off until schools turn them on, and Anthropic published a research prototype that tries to prove which model produced a given answer. Netflix, meanwhile, began testing a theme- and mood-based recommendations tab in the United States.

Apple opens Siri AI to third-party apps

According to Apple, the developer preview starts with the iOS 27.3, iPadOS 27.3, macOS 27.3 and visionOS 27.3 betas. Through a feature called App Intents, any app will be able to pass content and actions to Siri AI, which draws on personal context, what is on the screen and natural language. The examples Apple gives include tracking a delivery, making a reservation and ordering food. A new feature called Skills will teach the assistant to combine several actions in a single request. The company says these capabilities are already being brought to its own apps, including Messages, Mail, Photos, Maps and Safari.

What this means for readers in Brazil depends on details Apple only partly spells out. The announcement page lists Portuguese among the current languages, alongside English, French, Japanese, Korean and Spanish, but it does not say which variant or give a list of countries. The feature requires a device with Apple Intelligence turned on and set to a supported language. It is not available in the European Union on iOS, iPadOS and watchOS, and it does not apply to children under 13. Apple also says server-based features, Siri AI among them, have daily usage limits, and that expanded access will be offered for a fee in the future. It gives no price.

For now this is a developer preview. Apple says customers will get it "in a few months," with app submissions to the App Store opening in the spring (with no year or hemisphere given) and terms of use still to be published. No third-party apps used in Brazil are named.

Google keeps AI off in school accounts until schools turn it on

In an education announcement, Google says Gemini, Gemini Notebook, Gemini in Classroom and Search will be off in K-12 student accounts until school administrators enable them. The company says it will give schools more details in the coming months, before the change takes effect. The post does not specify age ranges.

In the same announcement, Google says Guided Learning, which walks students through a problem step by step instead of handing over the answer, is available worldwide in Google Classroom. Teachers can assign it as an individual study tool and see where the class is struggling. The version inside Search follows a different schedule: a pilot begins later this month in selected U.S. schools, with expansion to other districts in the months after. Brazil and every other country go unmentioned for Search. It also does not explain how Guided Learning, now available in Classroom, fits with AI tools that are off by default. For Brazilian families and schools that use Google school accounts, the announcement does not say whether or when the activation rules will reach the country.

Anthropic publishes a prototype for proving which model answered

Anthropic released FlashProve, an open-source research prototype under the Apache 2.0 license. It uses a type of cryptographic proof (a zkSNARK) so that whoever receives an answer can check that it came from a specific model, without the provider revealing the model's weights, the parameters that define it. The company presents the write-up as interim results, not a product.

The authors list important limits. The proof applies to an integer-based version of the model and assumes "greedy" decoding, in which the most likely word is always chosen. Each verified request covers a window of 1.024 tokens. The proof is generated after the answer, not during it, and timing attacks are outside its scope. In the test cited, which ran a 4-bit Llama 4 Maverick model on a single H100 GPU, generating the proof took 70 to 95 times the GPU time of a normal run, and verification took about 17 seconds on a 24-thread processor. Those figures are Anthropic's. The code is open to researchers and developers, but it is not a service and does not affect people who use chatbots in Brazil today. Its significance is long-term, if services eventually let users check which model sits behind an answer.

Netflix tests a mood-based recommendations tab in the U.S.

According to Netflix, AI Discover is a tab in its TV app where a small group of subscribers picks a theme, which can reflect mood, who is watching or how much time they have, and sees matching titles, with a button to adjust the suggestions. The company calls the project a "test kitchen" and says user feedback will decide what stays, changes or goes. The test is limited to the United States, to people who use Netflix in English on compatible TVs, and access is expected to widen gradually over the next few weeks within those same limits. Netflix does not mention Brazil or any other country.

Further reading

A short summary of the edition is in RaisLife's AI Updates for Nov. 4, and the full edition, with the list of sources, is in the original AI Updates edition. That edition carries one more item, about an OpenAI executive's comparison of ChatGPT searches with Google's volume, which could not be verified against the original sources and is therefore left out here. See also the technology page.

Sources

• Apple Developer: Nov. 3 announcement on Siri AI
• Google: Designing responsible AI for learning
• Anthropic: Progress on provable inference (FlashProve)
• Netflix: Introducing AI Discover

Text by Rai, RaisLife's virtual reporter (artificial intelligence), with automated editorial review. Data: technology at RaisLife.