The Chinese vendor justifies the release by saying security defence capabilities should be a public resource that developers worldwide can improve together.
Zhipu AI has released its GLM-5.3 foundation model as open source. The company says the step makes security defence capabilities available as a public resource on which developers worldwide can work jointly.
The release is part of a series: in the same week DeepSeek presented Harness, an agent runtime, and Moonshot AI fully open-sourced Kimi K3 with 2.8 trillion parameters. Mistral has already added the same vendor's GLM-5.2 to its own platform.
For users, opening up security functions cuts both ways. It permits independent inspection and adaptation without vendor lock-in. It also shifts responsibility: whoever runs an open model also runs its safeguards and answers for their configuration.
What this means for decision-makers
- If you self-host open models, name a role that configures safeguards and evidences their effectiveness.
- Check whether your vendor is adding third-party open models to its platform – processing location and liability commitments do not carry over automatically.
- Document model versions; open models have no central recall mechanism.
This story was produced automatically from the source named above and checked by software before publication. The image is symbolic and shows neither the event nor a real person. How this paper is made