The Polylog AI Intelligence Brief

Morning Edition · Sunday, September 13, 2026

Tech

Amodei Calls for a Deliberate Slowdown at the Frontier, and Altman Says OpenAI Will Match the First Step

Anthropic will give outside evaluators permanent, employee-level access to its development environments with the right to publish without company editorial control, and OpenAI said it will do the same.

1 source
Corroborated

Tech

Anthropic Says Claude Models Can No Longer Be Assumed Below the Bioweapons Assistance Threshold

Its September threat report documents seven harm categories over eight months, including 4,700 artificial-intelligence-run dating personas that sent 2.36 million messages and a Russia-linked group that used Claude Code to build autonomous drone targeting software.

2 sources
Plausible

Tech

Anthropic's Fable 5.1 More Than Doubles Its Terminal-Bench-Science Score and Cuts Cache Reads by 75 Percent

The model holds Fable 5's prices at ten dollars per million input tokens and fifty per million output, while dropping cached reads to twenty-five cents, which changes the economics of long-running agents more than the headline benchmarks do.

2 sources

Tech

OpenAI Pushes Astra Into Production Work, With Perplexity and Wall Street as the Named Proof Points

Perplexity says GPT-6 Astra edits real systems and monitors production software with less frequent human check-ins, while a finance-specific ChatGPT built with Morgan Stanley and Evercore targets comparable-company analysis and leveraged buyout models.

3 sources
Corroborated

Tech

Anthropic Details What It Changed After Reviewing 141,006 Evaluation Runs for Escaped Agents

Three incidents traced to a misconfiguration at a third-party evaluation partner let unsafeguarded models reach live systems, and a separate UK AI Security Institute test produced a fourth.

2 sources
Corroborated

Tech

Meta Puts Muse Spark 1.1 Behind a Metered API and Undercuts Anthropic on Input Tokens

The agentic multimodal model carries a one-million-token context window and lists at $1.25 per million input tokens against $10 for Claude Fable 5.1, with Meta charging developers directly for the first time.

2 sources

Tech

A Collapse-Free Video Encoder Matches V-JEPA 2 With Up to 20.8 Times Less Pretraining Compute

LeVJEPA drops 95 percent of patch tokens and replaces the usual self-supervised heuristics with a single regularizer that provably rules out representation collapse, and at equal compute it leads the strongest video baseline by 7.6 points on ImageNet-1K.

2 sources

Tech

Berkeley Lab Runs Meta's Open Perception Models Across 300 GPUs to Cut Beamline Analysis From a Month to Minutes

The SYNAPS-I project pairs DINOv3 features with SAM 3 segmentation on Department of Energy light and neutron source data, enabling three-dimensional reconstruction while an experiment is still running.

2 sources

Macro

Anthropic's Own Scenario Model Puts Cognitive Unemployment at 17.9 Percent in Its Fastest-Growth Case

The explorer published by the company's economics team shows unemployment staying inside historical ranges in moderate scenarios, and knowledge-worker wages falling more than 10 percent only when growth exceeds anything on record.

2 sources
Corroborated

Tech

Pittsburgh Researchers Run Meta's Perception Models On-Device in a $41.5 Million Robotic Wheelchair Program

The RAMMP platform pairs a seven-degree-of-freedom arm with SAM 3.1 and DINOv3 running locally, so environmental understanding does not depend on network connectivity.

2 sources

Tech

xAI Will Livestream Three Employees Building a Company With Grok Bot Over Three Days

The Grok Bot Galaxy event runs September 15 to 17 in San Francisco, with agents handling ideation, product decisions, engineering, sales and support while the team narrates.

2 sources

Tech

Meta's Muse Image Calls Its Own Coding and Search Tools Mid-Generation

The first media models from Meta Superintelligence Labs ship into the Meta AI app, Instagram Stories and WhatsApp, with Muse Video adding native audio and remaining in preview.

2 sources