ChatGPT Wallet Pilot, Amazon Product Graph, llama.cpp 1.0, Bay $50M, GitHub Updates

ChatGPT Wallet Limited Release

On November 13, OpenAI announced ChatGPT Wallet in partnership with American Express, Mastercard, and Visa. The initial rollout is gradual within Dots, limited to U.S. ChatGPT Pro users aged 18 or older who hold eligible U.S.-issued bank cards.

In the disclosed flow, the user first reviews the merchant, product, amount, and payment card, then confirms with Face ID or fingerprint. The payment network generates a substitute credential for that specific transaction, which the agent uses to complete the purchase while the real card number remains protected. For example, if a user asks an agent to find and buy an item, the final payment details are still presented for explicit approval before submission.

OpenAI also indicated a future goal of supporting delegated payments within preset limits and rules, but the initial release requires per‑transaction confirmation.

Amazon's Internal World Product Graph Concept

Reuters reported on November 13 that Amazon internally discussed building a World Product Graph database. According to documents seen by Reuters, the graph would include products Amazon does not sell itself, enabling a shopping assistant to answer questions about model numbers, specifications, use cases, and product relationships.

As an illustration, knowing a camera's name returns many results, but determining whether a specific accessory fits requires distinguishing body versions, interfaces, and accessory specs — exactly the relational data the graph aims to organize. Amazon responded that the referenced materials are outdated and may not reflect current plans, so the report should be treated as a disclosed internal concept.

llama.cpp 1.0.0 Introduces Semantic Versioning for Public APIs

On November 13, the open‑source inference project llama.cpp and its underlying tensor library ggml both released v1.0.0. The project provides a foundation for running models locally on diverse hardware and now supports over 150 model architecture variants across desktop and mobile environments.

The key change for developers is the adoption of semantic versioning for public C and binary interfaces. Any incompatible change to those interfaces will be reflected in the major version number, giving teams integrating llama.cpp into desktop tools or mobile apps a clearer signal of when to assess integration impact. Actual model performance continues to vary with hardware, model size, and configuration.

Bay Emerges with $50M Funding for Pre‑Action Permission Checks

Security startup Bay exited stealth on November 10, disclosing $50M in total funding: a $37M Series A led by Battery Ventures and a previously unannounced $13M seed round.

Bay's product targets enterprise devices, cloud services, and SaaS applications. It evaluates the full context of a task before an operation executes, judging whether the action complies with organizational policy. For instance, an agent may read team documents under one rule set, while modifying permissions requires a different rule set. Disclosed customers include Micron, Peloton, ClickUp, and Wix. Performance claims regarding interception speed and deployment effectiveness are based on vendor disclosures.

GitHub Expands Code Scanning Configuration and Credential Detection

On November 13, GitHub added repository‑level AI Scan enablement settings to the REST API public preview. Administrators can now manage scanning configuration at both organization and repository levels; enabling the configuration makes scans available on demand using the appropriate quotas.

Separately, on November 12, GitHub's Secret Scanning gained a new detection rule for Cloudflare Access service credentials that are accidentally committed. The two updates serve different purposes: the first centralizes code‑scanning management, while the second broadens the range of known credential types that are automatically flagged.