Z.ai says sorry for slurping up your code, open sources ZCode

Z.ai’s ZCode coding tool was found uploading encrypted full user workspaces and project histories to Alibaba Cloud, with decryption keys controlled solely by Z.ai and no disclosed user opt-out. ZCode apologized, said the uploaded data was never used for model training, removed the Repo Wiki feature, and claimed third-party reviews by CAICT and NSFOCUS confirmed all previously uploaded data was deleted. The company open-sourced ZCode, but the researcher who exposed the issue said pre-patch upload code and commit history had been wiped, leaving transparency concerns.

Analysis

The investable read-through is primarily to Alibaba Cloud’s enterprise trust discount, not a material near-term revenue event for BABA. AI coding tools sit directly on proprietary source code and credentials; a perceived gap between stated permissions and actual data handling raises procurement friction, longer security reviews, and a bias toward self-hosted or VPC deployments. That can reduce the speed at which Chinese enterprises convert AI pilots into recurring cloud/API spend, where the strategic prize is workload migration rather than model-token revenue.

BABA faces a modest 1-3 month sentiment and sales-cycle risk because its cloud infrastructure was the alleged destination, even if it was not implicated in product design or access decisions. BIDU has less direct exposure but shares the sector-level risk: enterprise buyers may demand auditable repositories, customer-held encryption keys, and contractual data-deletion rights before expanding coding-agent deployments. The company’s assertion that data was not used for training is not independently sufficient to remove the commercial overhang; missing historical upload code and commit history weakens forensic confidence.

Over 6-18 months, this favors vendors offering deployable-on-premise/private-cloud architectures and independently verifiable security controls over centrally hosted developer agents. The contrarian view is that public-equity impact is likely overestimated: BABA’s cloud scale and diversified earnings make a single third-party application incident immaterial unless regulators establish Alibaba Cloud had inadequate controls or affected customers identify exposed IP. A clean third-party report with reproducible deletion evidence would likely close the immediate discount quickly.

Key Decisions for Investors

• No standalone directional BABA short on this event: maintain a watch item until evidence emerges of customer churn, a regulatory inquiry, or Alibaba Cloud management acknowledging incremental compliance/remediation costs. The thesis is falsified by a credible independent report plus unchanged cloud deal activity over the next 1-2 quarters.
• For existing BABA exposure, favor a 1-3 month hedge via limited-risk BABA put spreads rather than reducing strategic positions; target strikes around 8-12% below spot, sized to a downside scenario in which cloud growth guidance is cut. Exit the hedge if the audit is published without adverse findings or BABA reports stable enterprise-cloud bookings.
• Use any broad China AI multiple compression to prefer BIDU over BABA on a relative basis only if BIDU demonstrates customer-controlled deployment and data-governance terms in its next enterprise AI disclosure. The pair fails if BIDU’s AI Cloud growth decelerates more sharply than Alibaba Cloud or similar security concerns surface.
• Monitor Chinese cybersecurity and data-governance procurement beneficiaries, including 688561.SS (Qianxin) and 002268.SZ (Weining), for evidence that enterprises are increasing code-scanning, endpoint, and private-deployment budgets. Do not initiate solely on the incident; require order/backlog confirmation because state and enterprise procurement cycles can lag 2-4 quarters.