01 Chatbots push shopping into conversational workflows
OpenAI today described a push to make ChatGPT a more visual, commerce-capable assistant, introducing an Agentic Commerce Protocol to power product discovery, side-by-side comparisons and merchant integrations. The company’s blog emphasized visually immersive product displays and APIs for merchants to connect with the chatbot experience.
Coverage across outlets notes the broader competitive context: Google’s Gemini is expanding commerce features too — including partnerships that let its assistant purchase clothes from participating retailers — and both companies are experimenting with different degrees of direct purchasing inside chat interfaces.
At the same time, OpenAI appears to be retrenching on parts of its earlier commerce push: reports say the company is moving away from Instant Checkout, the feature that enabled purchases directly through the ChatGPT interface, even as it builds richer discovery and integration tools.
- Chatbots are becoming commerce platforms, not just search or assistants.
- OpenAI’s new protocol focuses on discovery and merchant integration rather than fully committing to always-on checkout.
- Competition with Google is accelerating product features but also producing different tactical choices about in-chat payments.
02 Anthropic lets Claude act on your computer with built-in guardrails
Anthropic updated Claude Code and Cowork to let the assistant perform tasks directly on a user’s machine by opening files, using browsers and running developer tools. The Verge reports the feature requires no setup and can operate even when you’re away from your computer, while Anthropic frames it as a productivity boost for developers and power users.
TechCrunch describes an ‘auto’ mode that reduces the number of user approvals required for task execution. Anthropic says it balances speed with safety by keeping guardrails and constraints in place rather than granting unrestricted control.
The change reflects a broader industry shift toward more agentic, autonomous tooling that executes actions on behalf of users — accompanied, in Anthropic’s case, by built-in limits intended to reduce unsafe or unintended behavior.
- AI tools are increasingly able to act on a user’s device, moving beyond suggestions to execution.
- Anthropic’s auto mode shortens the approval loop while claiming to preserve safeguards.
- Expect more productivity gains but also ongoing scrutiny of how autonomy and safety are balanced.
03 OpenAI publishes teen-safety policies and open-source tools for developers
OpenAI released prompt-based teen safety policies and the gpt-oss-safeguard tooling to help developers address age-specific risks in AI systems. The company positions these resources as a practical starting point so teams don’t have to build teen-safety measures from scratch.
TechCrunch frames the announcement as an attempt to make it easier for third-party builders to adopt consistent protections for younger users, providing concrete moderation patterns and policies that can be integrated into applications.
The move signals attention to downstream responsibility as models are embedded in consumer and developer products — delivering reusable policies and code to reduce common safety gaps for teen audiences.
- OpenAI is shipping reusable safety policies to help developers moderate teen-specific risks.
- gpt-oss-safeguard gives teams a prompt-based approach they can integrate rather than inventing their own rules.
- This is an example of platform providers pushing developer-facing safety toolkits into the ecosystem.
What moved around the edges
OpenAI lays out its safety approach for Sora’s video tools
OpenAI says it built Sora 2 and the Sora app with safety protections to address risks from advanced video and audio generation; TechCrunch reports the Sora app lacked sustained engagement and is being shut down as attention shifts away from an AI-only social feed.
OpenAI BlogArm unveils its first in-house CPU, bound for Meta’s AI data centers
Arm announced its first self-designed chip, the Arm AGI CPU, and Meta is an early customer; the chip is aimed at inference workloads for cloud AI processing.
The Verge AIDoss raises $55M to bring AI inventory management into ERP systems
Doss closed a $55 million Series B to expand its AI-powered inventory management that integrates with existing ERP systems, backed by investors including Madrona and Premji Invest.
TechCrunch AITalat launches a subscription-free, local-first AI meeting-notes app
Talat launched a local-first, subscription-free AI meeting notes app that processes data on users’ machines rather than in the cloud, positioning itself as a privacy-focused alternative to cloud-based note tools.
TechCrunch AIApple pushes deeper into model competition
Hark’s founder, a former Apple designer, says the company will design models, hardware and interfaces together to deliver a cohesive personal intelligence product that blends software and hardware design.
TechCrunch AI