---
title: "Reverse Interoperability: Li Wenlong's Frame for the Doubao On-Device Agent Fight"
author: "DCC Editorial"
published: 2025-12-12T03:00:00.000Z
url: https://datacompliancechina.com/posts/doubao-reverse-interoperability-on-device-agents/
description: "ByteDance's Doubao phone assistant — preinstalled at the device layer to operate other apps on a user's behalf — was met with pop-up blocks from WeChat and others citing security and risk-control. Li Wenlong (科技利维坦) argues the dispute is, at bottom, a question of how China's competition-law toolkit (反不正当竞争法 / 反垄断法) absorbs the idea of interoperability — and specifically what he calls 'reverse interoperability (反向互操作性)'. The classic interoperability problem is a platform refusing to open up, with antitrust used as a market remedy to force access. Doubao inverts it: interoperability is fully achieved at the device level, and the legal question becomes whether the law should restrict 'over-interoperation.' Li maps interoperability's journey from the Microsoft case through GDPR data portability and the DMA to the agent era, distinguishes the Doubao fight from the decade-old 3Q War, and predicts on-device-agent governance will look less like classic antitrust and more like the ex-ante, conditional-use compliance model emerging for AI training data. For overseas counsel: a structural read on the platform-access war that on-device AI agents are about to intensify."
tags: ["ai-governance", "ai-agents", "interoperability", "anti-unfair-competition", "platform-governance", "on-device-agents", "academic-commentary"]
laws_cited: ["genai-services-interim-measures"]
domains: ["data-economy", "ai-governance"]
account: "keji-leviathan"
original_title: "豆包手机助手说到底是个被动互联问题"
original_author: "李汶龙 (Li Wenlong)"
original_publication: "科技利维坦 WeChat Official Account"
original_url: "https://mp.weixin.qq.com/s/E6RjF63A3vqd4WW78mVlXQ"
source_language: "zh"
---
> *Editor's Note — DCC.*
>
> This brief summarises 《豆包手机助手说到底是个被动互联问题》by Li
> Wenlong (李汶龙) on the 科技利维坦 channel — a theory-building piece on
> the Doubao phone-assistant controversy. The dispute itself: ByteDance
> preinstalled, on ZTE/Nubia handsets, a system-level AI assistant that
> can *operate other apps for the user*; WeChat and others responded with
> pop-up limits and interruptions citing security and risk-control,
> opening a fight between phone makers and "super-apps" over permissions,
> security and "entry control (入口控制权)." Li's move is to refuse the
> easy analogies ("disruptor," "theft," "intimate digital companion") and
> reach for a precise legal anchor: interoperability, and its inversion.
> This is a competition-law and platform-governance piece rather than a
> pure data-protection one — DCC runs it because the on-device AI agent
> is about to make platform data-access the central battleground of the
> Chinese internet, and Li's "reverse interoperability" framing is the
> sharpest public attempt to name what is actually being fought over.
> Note: the analysis is Li's own theoretical proposal, not settled
> Chinese law.

## The dispute, stated precisely

Strip away the metaphors and the Doubao question is narrow, Li says:
**why is it fine for a consumer to perform a sequence of taps on their
own phone, but not fine to let the phone do it automatically?** Where,
exactly, is the harm — and how do existing legal concepts, principles
and precedents attach to the new business form?

Public commentary has reached for analogies that all fall short.
Incumbents call Doubao a "disruptor (搅局者)" — but disruption is not
*per se* bad for consumers or markets; historically every technical
advance disrupts. Others analogise to theft — but what is stolen? Can
"traffic" be stolen; does the criminal-law concept of theft migrate
here? Still others coin labels — "system-level dangerous permissions,"
"intimate digital companion," "super-app / entry point," "phone
controller." Each touches *some* facet, none is precise. Regulation, Li
argues, needs a conceptual anchor to identify the applicable legal
framework and to pin down the *impaired legal interest*. His candidate
is interoperability.

## What interoperability actually covers

In China's internet-policy history, "互联互通" flared briefly and was
then suppressed; the public version touched only the surface — can a
Taobao link open inside WeChat, can WeChat Pay be used inside Douyin.
Those are interoperability symptoms, but the concept is far richer.

At its core, interoperability is the capacity of different systems,
platforms, technologies or services to mutually recognise, read/write
and effectively interact — at the **protocol** level (compatible
standards) and the **data** level (system A can read or write system B's
data); on Urs Gasser's account it extends to human and institutional
layers as well. Its legal home has shifted with each technical era:
from network-comms and computer engineering before the internet; into
**antitrust and competition policy** as software platforms began to
compete (the Microsoft case as watershed); then, in the platform era,
into digital-ecosystem antitrust enforcement, the **GDPR's data-
portability right**, and the **DMA's** messaging- and social-media-
interoperability mandates. With agents, the meaning shifts again — from
protocol and data interoperability to *semantic* understanding,
invoking another service's core capabilities, and completing complex
tasks across ports.

Doubao introduces **function-level interoperability**: a platform lets a
third party reach into its core functions. The subtlety, Li notes, is
that ByteDance did not negotiate app-by-app permission; it partnered with
a *handset maker* to solve the problem at the **device layer** — so the
assistant achieves seamless cross-app operation without any app's
consent.

## "Reverse" interoperability

Here is Li's central inversion. The classic interoperability problem
assumes the technology is *not* interoperable — there is no capability
or no incentive to open up — and the law (typically antitrust) pushes to
*force* access; opening the ecosystem wall is a market remedy. Doubao is
the opposite. Interoperability is *fully achieved* at the ZTE device
layer. So the question flips: the technology "over-interoperates," and
the issue becomes whether the law should *restrict* it. Li calls this
**reverse interoperability (反向互操作性)** — borrowing the
interoperability concept to bring the dispute into legal language, while
recognising that the legal problem is qualitatively different.

It is tempting, he notes, to liken Doubao to the decade-old **3Q War**
(Tencent v. Qihoo). Both touch the boundary between system platform and
app service, and both involve competitive exclusion. But the 3Q War was
about tampering with a function to block ads (framed as unfair
competition); Doubao is about using high system-level permissions to
*bypass an app's risk controls* and so unsettle the app-centric
structure of mobile internet. You cannot simply "reverse" the classic
antitrust framework and apply it here.

## Governing "reverse interconnection"

From the antitrust and regulatory angle, interoperability is normally a
*structural competition remedy*: it reduces user lock-in, weakens the
market power that network effects confer, and lets competitors into a
concentrated market. In competition law it shows up as three questions:
(1) must a dominant firm provide a technical interface or data to
rivals; (2) does refusing interoperability constitute abuse of
dominance; (3) does *mandating* interoperability promote competition
while avoiding over-intervention in innovation. The settled lesson:
interoperability can promote competition, but forced interoperability —
which tends toward standard unification — can blunt the incentive to
innovate.

Because Doubao is a *reverse* problem, that logic does not transplant
directly. Li's prediction is that on-device-agent governance will
instead come to resemble how **copyright law is learning to govern AI
training data** — a shift from purely *ex-post* infringement findings
toward *ex-ante* lawful access plus conditional use, wrapped in
AI-compliance duties (transparency, training-data-source disclosure,
filtering). By analogy, an on-device agent would be allowed to
auto-execute operations only on conditions: it follows platform rules,
is identifiable and accountable, and apps express their permissions in a
standardised, machine-readable way.

The competition-law apparatus still has a role at the edges. Li flags
the abuse-of-dominance analysis and the **DMA gatekeeper** path — asking
which platforms or services might constitute an *essential facility* in
the AI era — and suggests that, on a footing of platform safety and
control, agents could be offered access on **FRAND-style** terms, with
interoperability restricted or suspended where there is sufficient
evidence of harm. Crucially, cross-device operation spawns new risks
(privacy among them) that demand stronger accountability —
**permission audits and logging** — so on-device-agent compliance will
not be merely contractual; it will carry distinct AI-compliance
requirements.

## Why overseas counsel should care

- **The platform-access war is restarting at the device layer.** The
  Doubao fight is the leading edge of a structural conflict that
  on-device AI agents will intensify. Clients on either side — super-apps
  defending risk-control, or agent/handset players seeking access — need
  a frame sharper than "disruptor" or "theft," and "reverse
  interoperability" is it.
- **Expect ex-ante, conditional access, not just litigation.** If Li is
  right, the governance model borrows from AI-training-data compliance:
  standardised permission expression, identifiability, accountability,
  audit logs — built in up front rather than fought out after the fact.
- **Accountability artefacts matter now.** Permission audits and logging
  for cross-device actions are the controls most likely to harden into
  obligations; agent builders should be generating them already.

This connects directly to DCC's coverage of China's emerging AI-agent
rules — the
[governance framework](/posts/ai-agent-rules-governance-framework/) and
[risk taxonomy](/posts/ai-agent-rules-risk-taxonomy/) — and to Li's
companion piece on the same Doubao assistant viewed through the
privacy/authorisation lens,
[AI agents and the limits of consent](/posts/ai-agents-and-the-limits-of-consent/).
His training-data analogy also tracks DCC's note on
[open-source AI training-data compliance](/posts/open-source-ai-training-data-compliance/).

## DCC sources

- Original: 李汶龙 (Li Wenlong), 《豆包手机助手说到底是个被动互联
  问题》, 科技利维坦 WeChat Official Account
  ([source](https://mp.weixin.qq.com/s/E6RjF63A3vqd4WW78mVlXQ)).
- Comparative reference points named by the author: the US *Microsoft*
  antitrust case; the EU GDPR data-portability right; the EU Digital
  Markets Act (gatekeepers, interoperability mandates); Urs Gasser's work
  on interoperability; and China's 3Q War (Tencent v. Qihoo).
- The forward-looking compliance analogy tracks China's
  [Generative AI Services Interim Measures](/laws/genai-services-interim-measures/)
  model of ex-ante, disclosure-based AI duties.

> This is an editorial summary, not a translation. The "reverse
> interoperability" framing and the governance prediction are Li
> Wenlong's analytical proposals, not settled Chinese law; any
> simplification or error of emphasis is DCC's. **Not legal advice.**