The Polylog AI Intelligence Brief

Morning Edition · Friday, September 11, 2026

Geopolitics

Anthropic Ties 151 Million Claude Exchanges to an Alleged Alibaba Distillation Campaign

The September threat report names DeepSeek, Moonshot AI and MiniMax in smaller efforts, and puts model extraction alongside cyber operations and weapons research as a named harm category.

4 sources
Plausible

Tech

OpenAI Opens the Codex Harness as a Managed Agents API

The public beta charges no platform fee beyond tokens and tools, and lets developers run agents in OpenAI sandboxes or on Cloudflare, DigitalOcean and Oracle infrastructure.

3 sources

Tech

OpenAI Pauses New ChatGPT Pro Sign-Ups Because Astra Demand Exceeds Its Compute

The 200-dollar tier places the heaviest load on OpenAI's systems, and existing subscribers keep access while the company adds capacity.

3 sources

Tech

Claude Fable 5.1 Posts Large Gains on Terminal and Automation Benchmarks

Anthropic more than doubled the model's Terminal-Bench-Science score to 52.6 percent and cut cache-read pricing by 75 percent while holding headline token rates flat.

3 sources

Tech

OpenAI and the General Services Administration Cut Federal ChatGPT License Fees to Zero

The 27-month agreement runs from October 2026 through 2028, adds half-price usage and discounted access to OpenAI's Daybreak security platform, and extends to state, local and tribal governments.

3 sources

Markets

OpenAI Aims GPT-6 Astra at Investment Banking Workflows

ChatGPT for Financial Services ships with direct connections to data from the London Stock Exchange Group (LSEG), Daloopa and PitchBook, built with Morgan Stanley and Evercore as design partners.

3 sources

Tech

d-Matrix Will Plug Its Inference Chips Into NVIDIA Racks Through NVLink Fusion

Up to 144 Raptor accelerators can share a single NVLink domain, with rack-integrated systems expected in the fourth quarter of 2027.

3 sources

Tech

Skild AI's S1 Learns Robot Tasks From One Video Without Weight Updates

The model reports 66 percent success on unseen tasks against 9 percent for an equivalent language-prompted policy, on jobs running up to ten minutes.

3 sources

Macro

Anthropic's Economic Scenario Model Puts Its Own Chief Executive's Forecast in the Tail

The extreme case has 2030 output 32.4 percent above the no-AI path while cognitive-worker unemployment reaches 17.9 percent, but most modeled scenarios keep unemployment inside its historical range.

3 sources
Corroborated

Tech

Tencent Releases a 1.5-Billion-Parameter Speech Model Under an MIT License

AuK handles text-to-speech, content and acoustic editing, denoising and source separation through one natural-language instruction interface, with weights on Hugging Face and ModelScope.

3 sources

Tech

Researchers Find Agent Skills Work Better as Subagents Than as Context Injections

A Cornell and Microsoft Research paper argues fresh context windows beat loading skill instructions into a growing main context, when skills expose clear input-output contracts.

2 sources

Tech

A Latent-Space Language Model Scales to 8.9 Billion Parameters and Beats OLMo-3 on Reasoning

NCP-ArchPreview adds concept-level prediction on top of next-token training and reports a 5.99-point gain on the grade-school math benchmark GSM8K.

2 sources