Google says Search can now handle the whole tangled request you were about to break into five searches. That’s a welcome place for an AI upgrade: something you already use, without another app to adopt.
Search gets better at complicated requests
Gemini 4 Hydrogen arrived in Google Search on November 12, available globally to all users. Google says the model better understands multi-part requests, highly specific requirements and even hard-to-understand typos.
The change goes beyond writing a longer answer. Google says Search now runs simultaneous searches across the different parts of a request, connects users to a broader range of websites, and produces more images, comparison tables and structured responses. It also promises better understanding across all supported AI Mode languages.
For work, this is worth a small test with research you already need to do. Take a comparison you would normally split across several searches, put the requirements into one request, then open the linked websites to check the details behind the answer. Google specifically says Hydrogen follows each part of a complicated request more precisely.
There’s no new tool to learn for that test. Google says Hydrogen is available in Search for all users globally.
Muse can make the call
Meta’s Muse can now call U.S. businesses on behalf of paid subscribers in the U.S., expanding a feature previously limited to selected early users.
Meta chief AI officer Alexandr Wang names cable companies, the DMV and restaurants without online reservations as examples. If you already pay for Muse and are in the U.S., this gives you another task to hand over. An ordinary reservation seems a reasonable place to start.
There is still a person on the other end who has to accept the call. TechCrunch’s Sarah Perez notes that businesses and receptionists sometimes hang up when they recognize an AI agent. Muse plans start at $20 a month for 500 million tokens per week; the $100 plan includes 3 billion.
Notion chats exposed through a provider
Fireworks AI confirmed a breach that exposed customer data. Notion says the stolen data included some of its Notion AI chat transcripts and that around 1% of its customers were affected.
Fireworks says an attacker breached its cloud provider and accessed a development system. Notion says the stolen transcripts came from its “Insights” feature and included inputs and outputs written to the chat. It adds that much of the data could not be attributed to a particular customer.
For teams deciding what work material belongs in an AI chat, this is a concrete reminder that the provider chain matters alongside the app’s own security. Notion says it disconnected its systems from Fireworks, deleted the data stored in Notion’s systems and stopped connectivity between the two on November 3.
Also
- Risky shopping links. Podqi’s test of 6,795 chatbot answers found risky retail sites mentioned or linked in 4% of Gemini answers, 8% of ChatGPT answers and 15% of Perplexity answers, with mentions including warnings.
- An already-known Maya site. Pietro Miotti acknowledged that his 48 AI agents missed prior research documenting the Maya site he had suggested was potentially unmapped.
- Granite security preview. IBM released two experimental models designed to distinguish instructions from data containing malicious commands, explicitly marking them unsuitable for production.
Pass it on
