Daily AI RoundupsWednesday, 30 September 2026

Anthropic reports that Zhipu AI’s GLM-5.3 achieved control-flow hijacks in 4 of 100 exploit-benchmark tasks. OpenAI begins rolling out Dots, its always-on agents, and introduces GPT-6.1 Sol at one-fifth of Astra’s standard API token prices. DeepSeek released an Ascend-compatible kernel library for Huawei chips. A US appeals court rejected Ross Intelligence’s fair-use defense over Thomson Reuters Westlaw content, while a voluntary White House accord calls for external AI safety audits. Rachel Webb and Sebastian Raschka add practitioner perspectives on mathematics and text classification.

• AI security
• AI agents
• Model releases
• AI infrastructure
• Copyright
• Research readings

Stories in this roundup

1. 01
   Anthropic finds GLM-5.3 achieved control-flow hijacks in 4 of 100 exploit-benchmark tasksAnthropic
2. 02
   OpenAI begins rolling out GPT-6 Astra-powered Dots agents with cloud computers and plugin accessOpenAI
3. 03
   OpenAI introduces GPT-6.1 Sol at one-fifth of Astra’s standard API token pricesOpenAI
4. 04
   DeepSeek releases DeepGEMM-Ascend, an API-compatible kernel library for Huawei’s Ascend chipsDeepSeek
5. 05
   US appeals court rejects Ross Intelligence’s fair-use defense over Westlaw-trained legal searchReuters
6. 06
   Trump and AI executives sign voluntary safety accord calling for external audits and board oversightReuters
7. 07
   OpenAI rolls out reusable Codex cloud environments for coding across devicesOpenAI
8. 08
   Rachel Webb says AI speeds her mathematical research without changing what she finds interestingTerence Tao’s blog
9. 09
   Sebastian Raschka argues Jev raises the bar for fine-tuning specialist text classifiersSebastian Raschka