Ant’s LingGuang 2.0 adds ongoing cloud tasks in China rollout

Ant Group announced LingGuang 2.0 on October 27, adding ongoing task management to its multimodal AI assistant. The company says the upgrade is rolling out in China on iOS, Android and the LingGuang website. The announcement does not establish a worldwide rollout.

The practical change is what can happen between conversations. Ant says LingGuang runs on independent cloud instances and can keep following a task as information or its status changes, rather than ending its work when one chat session closes. Cross‑session memory, longer‑running task scheduling and tool calling underpin that approach, according to the company.

This is not described as unlimited standing authority. Ant says the agent draws on connected information and services with explicit user authorization, then initiates analysis, follows up and alerts the user when attention is needed.

The [release’s Alipay example](https://via.tt.se/pressmeddelande/4616995/ant‑group‑upgrades‑lingguang‑to‑20‑turning‑its‑ai‑assistant‑into‑a‑proactive‑personal‑agent?lang=en&publisherId=259167) describes reviewing authorized spending records, identifying potentially forgotten auto‑renewing subscriptions, analyzing spending patterns and offering savings recommendations. Those are analysis and recommendation tasks; the announcement does not establish automatic subscription cancellation or payment authority. For completed flight or hotel bookings, Ant describes monitoring price changes and notifying users when a lower price becomes available. A notification is distinct from rebooking a trip, obtaining a refund or guaranteeing the best available price. The company also lists built‑in capabilities involving Alipay billing, membership and travel services, email, DingTalk, Lark and Mi Home. Those connections extend the proposed workflow beyond a single chat, but the release does not supply a detailed permission or account‑eligibility matrix for each service.

LingGuang 2.0 incorporates Ant’s [APASS account and security management service](https://via.tt.se/pressmeddelande/4616995/ant‑group‑upgrades‑lingguang‑to‑20‑turning‑its‑ai‑assistant‑into‑a‑proactive‑personal‑agent?lang=en&publisherId=259167). The company says its design separates authorization keys from agents and holds the keys in escrow, allowing tasks to run without the agent directly accessing credentials. It also says users can review, adjust or revoke authorizations at any time. These are Ant’s descriptions of the authorization architecture, rather than an independent test of credential handling or permission enforcement. The relevant distinction for a continuing task is between what the agent keeps tracking and which connected services it has been authorized to use. The rollout announcement supplies examples of that design, without demonstrating universal autonomy, guaranteed savings or reliable completion of every task. Ant A Space in Hangzhou, photographed in December 2021. Archive company context, not a LingGuang 2.0 interface, output or October 2026 launch photograph. Photo: Siyuwj, via [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Ant_A_Space,_Hangzhou,_2021‑12‑02.jpg), [CC BY‑SA 4.0](https://creativecommons.org/licenses/by‑sa/4.0/). Cropped; image adaptations retain the same licence.