AI news in context

Google's conversational search experience can now connect to select third-party apps to complete tasks like grocery shopping, design, and playlist curation.

• Google AI Mode now supports direct app integrations, starting with Instacart, Canva, and YouTube.
• Users can complete tasks like adding groceries to a cart, browsing design templates, and saving playlists without leaving the AI Mode experience.

The AI lab known for software entered the physical goods market this week with a productivity keyboard and a rubber basketball that urges you to step away from the screen.

• OpenAI released its first hardware: a $230 mini keyboard described as a 'command center for agentic work.'
• The company also listed a $70 ChatGPT-branded rubber basketball tied to a 'Pause. Play. Prompt.' campaign encouraging breaks from screens.

Andrew Dai's Elorian, still pre-product, secured one of the most aggressive seed-stage valuations in recent AI history by betting that visual reasoning is the next frontier.

• Andrew Dai, former Google DeepMind researcher, raised a $55M seed round at a $300M valuation for Elorian before launching a product.
• Dai's prior research helped inform the development of ChatGPT; he spent over a decade at DeepMind.

Alexandre LeBrun, who runs Yann LeCun's world-model startup, says the industry's favorite buzzwords miss the point of what AI actually needs.

• Alexandre LeBrun, CEO of AMI Labs, publicly dismisses the terms 'AGI' and 'superintelligence' for his company's work.
• AMI Labs is building world models — AI systems designed to understand the physical world, not just predict text tokens.

1Password's new Claude integration uses a zero-exposure architecture to let AI agents complete browser tasks that require logins—without credentials entering the model or its memory.

• 1Password for Claude uses a zero-exposure architecture: credentials never enter the model or its memory.
• When Claude needs to sign in, 1Password prompts the user with biometric approval and injects the credential directly into the page.

A new browser integration gives Claude secure access to your stored credentials so it can handle multi-step tasks like booking travel, all without exposing secrets to the model.

• 1Password launched a browser integration called 1Password for Claude that lets the AI agent authenticate into websites on a user's behalf.
• The feature uses a zero-exposure framework: credentials never reach the model, its memory, or Anthropic's systems—they are injected directly into the target site.

Under the Digital Markets Act, Google must give competing AI assistants and search engines comparable access to Android features and some Search data, with deadlines stretching into 2027.

• The EU ordered Google to give rival AI assistants and search engines greater access to Android and Google Search under the Digital Markets Act.
• Google must begin sharing search data by January 2027 and implement Android changes by July 2027.

Ultrahuman alum Apoorv Shankar's new startup wants wearables to act on your behalf, with a pilot device coming in weeks.

• Aina raised $5.5M led by Redstart Labs and 360 ONE, with participation from MIXI, Antler, and Blume Founders Fund.
• Founder Apoorv Shankar previously served as VP of Hardware at Ultrahuman and founded LazyCo, which Ultrahuman acquired.

The company built a dedicated detection system, will manually review flagged chats, and is working with over 75 clinicians to improve crisis responses as scrutiny over chatbot safety for minors intensifies.

• Meta will notify parents using Instagram supervision tools if their teen discusses suicide or self-harm with the Meta AI chatbot.
• A dedicated AI system flags relevant conversations, but all flagged chats are manually reviewed before a parent alert is sent; Meta will err on the side of caution when intent is ambiguous.