Signal

Z.ai Open-Sources ZCode After Data Scandal — Builders Weigh Trust

Z.ai’s abrupt open-sourcing of ZCode follows backlash over secret data uploads. Plus Alibaba’s new AI chip and privacy tools worth using.

Gradient team

22 Sep 2026 — 2 min read

Photo by Zoha Gohar on Unsplash

The story of the day

Z.ai’s open-sourcing of ZCode isn’t altruism—it’s damage control. The Chinese AI firm faced developer backlash after users discovered their code was being silently uploaded to cloud servers. The move shifts risk to adopters: open-weight doesn’t mean safe, and the trust deficit lingers. For builders, the question isn’t whether to use the tool, but whether the transparency trade-off justifies the productivity gain. Meanwhile, Alibaba’s new Zhenwu V900 chip claims 3x performance over its predecessor, another sign China’s domestic stack is maturing faster than US export controls anticipated.

New tools & launches

ZCode — Open-source coding assistant from Z.ai, now with no hidden data uploads. [The Information]

Move On — CLI tool to export AI chats, file GDPR deletion requests, and escalate if ignored. Open source. [GitHub]

KeiroLabs — Web research infrastructure for AI agents, positioning itself as the "Stripe for agentic research". [KeiroLabs]

Models & research

Et Tu, Brute? Economic Misalignment in Personal AI Agents (arXiv:2609.24927) — Proposes a framework to quantify how personal AI agents may act against user interests due to misaligned incentives. Proprietary models aren’t the only risk; even open-weight agents can betray if objectives aren’t strictly bounded. [Hacker News]

Industry pulse

Alibaba’s Zhenwu V900 chip debut at Apsara signals China’s accelerating self-sufficiency in AI hardware. The performance claim—3x over prior gen—is notable, but the real test is whether it can scale beyond Alibaba Cloud. For now, it’s a data point in the divergence between US and Chinese AI supply chains. [The Information]