Augur Dispatch

Chain of evidence

Evidence for 2026-08-06

This frozen page shows Augur's claims and source links for one sent dispatch. Stored spot-checks appear only where the frozen edition supports them; absence is not presented as verification.

As of:

Bundle identity: evidence-bundle-v1-0d190d90427ad4433b136f6d15577d22bffb346df2b740786c900da0c39f30a9

Format: evidence-bundle-v1 · 33 claims

Assertion 1

The seed money is led by Radical Ventures and Khosla Ventures, with participation from Lightspeed, Kleiner Perkins, Doerr Capital, and Alphabet itself Latent Space.

Assertion status: No spot-check verdict is published for this assertion.

Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le departed Google DeepMind to cofound a new startup named Discovery Loop.

Claim 42904 Label: fact Provenance: primary Recorded

Latent Space

No stored spot-check names this claim in this edition.

Discovery Loop is structured as a Public Benefit Corporation focused on automating machine learning, science, and engineering through autoresearch loops.

Claim 42905 Label: fact Provenance: primary Recorded

Latent Space

No stored spot-check names this claim in this edition.

Radical Ventures and Khosla Ventures are leading the seed funding round for Discovery Loop, with participation from Lightspeed, Kleiner Perkins, Doerr Capital, and Alphabet.

Claim 42906 Label: fact Provenance: primary Recorded

Latent Space

No stored spot-check names this claim in this edition.

Assertion 2

But a year of evaluations shows agents making progress mainly on tasks where success is easy to check, and agents still cannot do open-ended AI research AI as Normal Technology.

Assertion status: No spot-check verdict is published for this assertion.

Leading AI labs aim for recursive self-improvement (RSI), defined as the automation of AI research using AI agents.

Claim 42703 Label: fact Provenance: primary Recorded

AI as Normal Technology

No stored spot-check names this claim in this edition.

Forecasts of explosive AI progress are underpinned by the concept of recursive self-improvement.

Claim 42704 Label: fact Provenance: primary Recorded

AI as Normal Technology

No stored spot-check names this claim in this edition.

Over the last year, many evaluations found that AI agents can make progress on tasks where success is easily verifiable.

Claim 42705 Label: fact Provenance: primary Recorded

AI as Normal Technology

No stored spot-check names this claim in this edition.

Assertion 3

Don't Worry About the Vase argues the frontier labs themselves are what it calls "ASI pilled," meaning convinced that artificial superintelligence, AI beyond human ability, is genuinely coming, and that a large share of their employees hold that view Don't Worry About the Vase.

Assertion status: No spot-check verdict is published for this assertion.

The author asserts that a large percentage of employees at frontier AI labs hold the 'ASI pilled' view.

Claim 42786 Label: fact Provenance: primary Recorded

Don't Worry About the Vase

No stored spot-check names this claim in this edition.

The author asserts that the frontier AI labs themselves are 'ASI pilled.'

Claim 42787 Label: fact Provenance: primary Recorded

Don't Worry About the Vase

No stored spot-check names this claim in this edition.

The author asserts that in fair or open competition, AI will reliably outcompete humans in all senses, including for resources.

Claim 42790 Label: forecast Provenance: primary Recorded

Don't Worry About the Vase

No stored spot-check names this claim in this edition.

Assertion 4

government to help pace it Don't Worry About the Vase.

Assertion status: No spot-check verdict is published for this assertion.

Over 1,290 employees at frontier labs signed an open letter warning that the industry is close to automating AI research and racing ahead faster than it can handle.

Claim 39821 Label: fact Provenance: primary Recorded

Don't Worry About the Vase

No stored spot-check names this claim in this edition.

The open letter signed by over 1,290 frontier lab employees requests that the U.S. government support an international effort to develop tools to deliberately pace the frontier of automated AI development.

Claim 39822 Label: fact Provenance: primary Recorded

Don't Worry About the Vase

No stored spot-check names this claim in this edition.

Assertion 5

Software supply chain attacks have been climbing for weeks, from a compromise of more than 30 Red Hat npm packages that Wiz researchers identified as the Miasma attack to today's escalation Wiz Research.

Assertion status: No spot-check verdict is published for this assertion.

Wiz researchers identified the Miasma supply chain attack, which compromised over 30 Red Hat Cloud Services npm packages.

Claim 39100 Label: fact Provenance: primary Recorded

Wiz Research

No stored spot-check names this claim in this edition.

Assertion 6

Microsoft identified an npm attack, npm being the main registry of reusable JavaScript code, affecting more than 400 packages, with malicious releases carrying what it calls a Mini Shai-Hulud variant: a self-propagating worm that steals credentials and spreads through a heavily obfuscated, meaning deliberately hard-to-read, JavaScript payload Microsoft Security Blog.

Assertion status: No spot-check verdict is published for this assertion.

Microsoft Threat Intelligence identified a large-scale npm supply chain attack affecting more than 400 packages.

Claim 42582 Label: fact Provenance: primary Recorded

Microsoft Security Blog

No stored spot-check names this claim in this edition.

The malicious releases contain a Mini Shai-Hulud variant, which is a self-propagating credential-stealing worm.

Claim 42583 Label: fact Provenance: primary Recorded

Microsoft Security Blog

No stored spot-check names this claim in this edition.

The worm is delivered through a large, heavily obfuscated Bun-based JavaScript payload.

Claim 42584 Label: fact Provenance: primary Recorded

Microsoft Security Blog

No stored spot-check names this claim in this edition.

Assertion 7

Nate Jones documents a Deloitte report that only one academic bothered to check, at a cost of A$97,587, alongside a government department, an open-source maintainer, and a federal judge who each paid real money or hours for AI content nobody verified Nate Jones.

Assertion status: No spot-check verdict is published for this assertion.

A government department incurred financial costs due to AI-generated content that had not been verified for accuracy.

Claim 42695 Label: fact Provenance: primary Recorded

Nate Jones

No stored spot-check names this claim in this edition.

An open-source maintainer incurred costs associated with AI-generated content that lacked human verification.

Claim 42696 Label: fact Provenance: primary Recorded

Nate Jones

No stored spot-check names this claim in this edition.

A federal judge incurred costs, including invoices and hours, due to AI-generated content that was not verified.

Claim 42697 Label: fact Provenance: primary Recorded

Nate Jones

No stored spot-check names this claim in this edition.

Assertion 8

OpenAI cut GPT-5.6 prices on July 30, dropping Luna by 80% and Terra by 20%, AWS matched those reductions on Bedrock, its rent-a-model service, effective the same date, and Bedrock's GPT-5.6 rates already track OpenAI's own first-party pricing Simon Willison's Weblog AWS News Blog AWS Machine Learning Blog.

Assertion status: No spot-check verdict is published for this assertion.

OpenAI's GPT-5.6 Sol, Terra, and Luna models are generally available on Amazon Bedrock as of July 13, 2026.

Claim 22834 Label: fact Provenance: primary Recorded

AWS Machine Learning Blog

No stored spot-check names this claim in this edition.

Pricing for GPT-5.6 models on Amazon Bedrock matches OpenAI's first-party rates.

Claim 22839 Label: fact Provenance: primary Recorded

AWS Machine Learning Blog

No stored spot-check names this claim in this edition.

OpenAI reduced the price of GPT-5.6 Terra by 20%.

Claim 40193 Label: fact Provenance: primary Recorded

Simon Willison's Weblog

No stored spot-check names this claim in this edition.

OpenAI reduced the price of GPT-5.6 Luna by 80%.

Claim 40194 Label: fact Provenance: primary Recorded

Simon Willison's Weblog

No stored spot-check names this claim in this edition.

AWS announced price reductions for OpenAI's GPT-5.6 models on Amazon Bedrock effective July 30, 2026.

Claim 42412 Label: fact Provenance: primary Recorded

AWS News Blog - Artificial Intelligence

No stored spot-check names this claim in this edition.

On-demand inference prices for the GPT-5.6 Luna model on Amazon Bedrock were reduced by 80% effective July 30, 2026.

Claim 42413 Label: fact Provenance: primary Recorded

AWS News Blog - Artificial Intelligence

No stored spot-check names this claim in this edition.

On-demand inference prices for the GPT-5.6 Terra model on Amazon Bedrock were reduced by 20% effective July 30, 2026.

Claim 42414 Label: fact Provenance: primary Recorded

AWS News Blog - Artificial Intelligence

No stored spot-check names this claim in this edition.

Assertion 9

Redo any cost comparison made before the cut, and factor in the 90% discount on cached input, meaning prompt text the service has already processed and can reuse AWS Machine Learning Blog.

Assertion status: No spot-check verdict is published for this assertion.

OpenAI GPT-5.6 Sol, Terra, and Luna models are now generally available on Amazon Bedrock.

Claim 39948 Label: fact Provenance: primary Recorded

AWS Machine Learning Blog

No stored spot-check names this claim in this edition.

Cached input tokens for GPT-5.6 on Amazon Bedrock are billed at a 90 percent discount compared to uncached input.

Claim 39958 Label: fact Provenance: primary Recorded

AWS Machine Learning Blog

No stored spot-check names this claim in this edition.

Assertion 10

- Discovery Loop's first public output will show whether autoresearch loops can beat the verifiable-task ceiling that current agents hit. Latent Space

Assertion status: No spot-check verdict is published for this assertion.

Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le departed Google DeepMind to cofound a new startup named Discovery Loop.

Claim 42904 Label: fact Provenance: primary Recorded

Latent Space

No stored spot-check names this claim in this edition.

Discovery Loop is structured as a Public Benefit Corporation focused on automating machine learning, science, and engineering through autoresearch loops.

Claim 42905 Label: fact Provenance: primary Recorded

Latent Space

No stored spot-check names this claim in this edition.

Radical Ventures and Khosla Ventures are leading the seed funding round for Discovery Loop, with participation from Lightspeed, Kleiner Perkins, Doerr Capital, and Alphabet.

Claim 42906 Label: fact Provenance: primary Recorded

Latent Space

No stored spot-check names this claim in this edition.

Assertion 11

- Microsoft's ChainDrop investigation could widen past 400 packages, and the registry's cleanup speed will set the tone for the next worm. Microsoft Security Blog

Assertion status: No spot-check verdict is published for this assertion.

Microsoft Threat Intelligence identified a large-scale npm supply chain attack affecting more than 400 packages.

Claim 42582 Label: fact Provenance: primary Recorded

Microsoft Security Blog

No stored spot-check names this claim in this edition.

The malicious releases contain a Mini Shai-Hulud variant, which is a self-propagating credential-stealing worm.

Claim 42583 Label: fact Provenance: primary Recorded

Microsoft Security Blog

No stored spot-check names this claim in this edition.

The worm is delivered through a large, heavily obfuscated Bun-based JavaScript payload.

Claim 42584 Label: fact Provenance: primary Recorded

Microsoft Security Blog

No stored spot-check names this claim in this edition.

Assertion 12

- AWS matching OpenAI's rates suggests the next GPT price change hits Bedrock immediately; the next model release tests that pattern. AWS News Blog

Assertion status: No spot-check verdict is published for this assertion.

AWS announced price reductions for OpenAI's GPT-5.6 models on Amazon Bedrock effective July 30, 2026.

Claim 42412 Label: fact Provenance: primary Recorded

AWS News Blog - Artificial Intelligence

No stored spot-check names this claim in this edition.

On-demand inference prices for the GPT-5.6 Luna model on Amazon Bedrock were reduced by 80% effective July 30, 2026.

Claim 42413 Label: fact Provenance: primary Recorded

AWS News Blog - Artificial Intelligence

No stored spot-check names this claim in this edition.

On-demand inference prices for the GPT-5.6 Terra model on Amazon Bedrock were reduced by 20% effective July 30, 2026.

Claim 42414 Label: fact Provenance: primary Recorded

AWS News Blog - Artificial Intelligence

No stored spot-check names this claim in this edition.

Assertion 13

- Simon Willison one-shotted a complete 3D browser game with Claude Fable 5, a marker of where coding agents sit on checkable, self-contained builds. Simon Willison's Weblog

Assertion status: No spot-check verdict is published for this assertion.

Simon Willison used Claude Fable 5 running in Claude Code for web to build a complete 3D browser game called "Raccoon Heist" based on a 2024 tweet concept.

Claim 42850 Label: fact Provenance: primary Recorded

Simon Willison's Weblog

No stored spot-check names this claim in this edition.

The "Raccoon Heist" game project is available at the GitHub repository simonw/raccoon-heist.

Claim 42851 Label: fact Provenance: primary Recorded

Simon Willison's Weblog

No stored spot-check names this claim in this edition.

Claude Fable 5 generated game textures using the OpenAI gpt-image-2 model API.

Claim 42852 Label: fact Provenance: primary Recorded

Simon Willison's Weblog

No stored spot-check names this claim in this edition.

Assertion 14

- OpenHands made GLM 5.2 the default model for its Canvas feature, a small sign that the default model choices inside agent tools are shifting. OpenHands Releases

Assertion status: No spot-check verdict is published for this assertion.

The OpenHands software version 1.10.0 was released on August 5, 2026.

Claim 42765 Label: fact Provenance: primary Recorded

OpenHands Releases

No stored spot-check names this claim in this edition.

The GLM 5.2 model is set as the default model for the Canvas feature in OpenHands version 1.10.0.

Claim 42766 Label: fact Provenance: primary Recorded

OpenHands Releases

No stored spot-check names this claim in this edition.

OpenHands version 1.10.0 includes a feature allowing users to export activity logs.

Claim 42767 Label: fact Provenance: primary Recorded

OpenHands Releases

No stored spot-check names this claim in this edition.

Assertion 15

Discovery Loop's structure and investor list come from a single newsletter roundup rather than a company filing, so the seed round details remain directional Latent Space.

Assertion status: No spot-check verdict is published for this assertion.

Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le departed Google DeepMind to cofound a new startup named Discovery Loop.

Claim 42904 Label: fact Provenance: primary Recorded

Latent Space

No stored spot-check names this claim in this edition.

Discovery Loop is structured as a Public Benefit Corporation focused on automating machine learning, science, and engineering through autoresearch loops.

Claim 42905 Label: fact Provenance: primary Recorded

Latent Space

No stored spot-check names this claim in this edition.

Radical Ventures and Khosla Ventures are leading the seed funding round for Discovery Loop, with participation from Lightspeed, Kleiner Perkins, Doerr Capital, and Alphabet.

Claim 42906 Label: fact Provenance: primary Recorded

Latent Space

No stored spot-check names this claim in this edition.