This observation covers important technical changes published and verifiable around the period from 20:00 on 10 September to 20:00 on 11 September 2026 in Australia/Sydney time. Three items are included.
OpenAI opens the Agents API in public beta
On 10 September, OpenAI released the Agents API and opened it to all developers as a public beta. Compared with building model calls, tool execution, context management and multi-agent orchestration separately, the new API exposes part of the agent runtime used by Codex as a managed service. Developers can specify the model, tools and execution environment in a session, choosing an OpenAI-hosted sandbox, their own infrastructure or a partner environment. OpenAI also documents long-session context compaction, on-demand tool search, programmatic tool calling and parallel subagents.
The service is currently a public beta rather than generally available. OpenAI says it is available to all developers and carries no separate Agents API platform fee; users continue to pay for the model tokens and tools they use. Performance figures quoted from participating customers remain self-reported and are not treated here as independent validation. There is not yet a common public third-party reproduction covering long-running reliability, isolation strength across sandbox choices, or complex production workloads.
Anthropic publishes its September 2026 AI misuse threat-intelligence report
On 10 September, Anthropic published “Detecting and countering misuse of AI: September 2026”. The report covers cases its Threat Intelligence team says it identified and disrupted between December 2025 and August 2026 across cyber operations, surveillance, influence operations, scams and fraud, biological misuse, conventional weapons development and illicit distillation. Anthropic says the cases mainly involved Claude Haiku, Sonnet and Opus. With one illicit-distillation exception, the reported misuse did not involve the Fable or Mythos model families.
Compared with earlier incident-by-incident disclosures, the new material is a consolidated eight-month report spanning several misuse categories. Anthropic says it banned accounts, strengthened safeguards and, where appropriate, shared intelligence with authorities or industry partners. The Associated Press separately reported the publication and several of the cyber, biological-research and other misuse cases, but the underlying account records, operational logs and attribution remain primarily Anthropic’s internal threat-intelligence evidence and cannot be fully reproduced externally. This edition therefore treats the detailed claims as developer-reported findings and confirmed disclosures, rather than independently verified attribution or scale measurements.
d-Matrix plans Raptor XPU integration with NVIDIA NVLink Fusion
On 10 September, d-Matrix and NVIDIA announced that d-Matrix’s next-generation Raptor inference XPU will adopt NVIDIA NVLink Fusion and is planned to connect into NVLink scale-up networking, Spectrum-X networking and the MGX rack architecture, alongside NVIDIA GPU systems. Relative to d-Matrix’s earlier standalone inference hardware, the new technical arrangement places its next-generation accelerator inside NVIDIA’s existing interconnect, networking, rack, power and liquid-cooling ecosystem rather than requiring a separate full rack architecture for Raptor.
This remains an integration and deployment plan, not a shipping production system. Reuters reported that the final Raptor design is expected by the end of 2026 and that integrated systems are targeted for 2027. NVIDIA’s material identifies the intended components, but no completed rack-level validation results or independent performance reproduction have been published. Claims about low latency and deployment efficiency therefore remain vendor targets; what is confirmed at this stage is the integration path and the scope of the collaboration.
Sources
OpenAI, “Introducing the Agents API”, 10 September 2026, https://openai.com/index/introducing-the-agents-api/
Anthropic, “Detecting and countering misuse of AI: September 2026”, 10 September 2026, https://www.anthropic.com/threat-intelligence-report-september-2026
Associated Press, “Anthropic says it blocked misuse of its AI that could have supported biological weapons”, 11 September 2026, https://apnews.com/article/00266dca90e4f8853f669648998d3bda
NVIDIA, “d-Matrix Adopts NVIDIA NVLink Fusion for Rack-Scale XPU Deployment”, 10 September 2026, https://blogs.nvidia.com/blog/d-matrix-nvlink-fusion/
Reuters, “Chip startup d-Matrix to use Nvidia chip-linking tech in AI servers”, 10 September 2026, https://www.reuters.com/business/media-telecom/chip-startup-d-matrix-use-nvidia-chip-linking-tech-ai-servers-2026-09-10/
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.