OpenAI expands Daybreak, Sakana releases Fugu, GLM-5.2 gains traction

This week's AI news highlights OpenAI's expansion of its cybersecurity initiatives, Sakana AI's release of an orchestration model called Fugu, and the growing adoption of the open-weight GLM-5.2 model.

• OpenAI expanded its Daybreak program with a Codex Security plugin and GPT-5.5-Cyber model, shifting focus to closed-loop patch generation with human review across major projects like cURL and Python.
• Sakana AI introduced Fugu, an orchestration system that learns model selection and verification across multiple frontier models, though it faced criticism for opaque baselines and missing cost accounting.
• GLM-5.2 emerged as a leading open-weight model for agentic work, ranking third on GDPval-AA and demonstrating robustness in real-world bug fixing tasks at a lower cost than proprietary alternatives.

These developments reflect a widening gap between model capability reporting and governance criteria, while shifting industry focus toward orchestration layers and the practical utility of open-source models.