Augur Dispatch

Chain of evidence

Evidence for 2026-09-10

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-v1-3ddc15dd0efb339c97192ff1fcd4f66d902843413ea4157e52096596335a7c03

Format: evidence-bundle-v1 · 49 claims

Assertion 1

Latent Space OpenAI News

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

OpenAI shared an AI-generated solution to the Navier–Stokes Millennium Prize Problem, including a writeup and a formal proof in Lean.

Claim 58202 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

OpenAI News

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

Anthropic announced that Claude completed the first fully computer-checked proof of Fermat's Last Theorem in Lean, producing 13 million lines of code and 29,500 supporting theorems over 11 days.

Claim 58721 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Latent Space

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

Assertion 2

Trail of Bits

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

Anthropic announced a formalization of Fermat's Last Theorem using 13 million lines of Lean code.

Claim 59001 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Trail of Bits

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

A Lean bug allows a proof of Fermat's Last Theorem to be checked as correct by exploiting a discrepancy between Lean's logical definition and its compiled native code for string extraction at large positions.

Claim 59002 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Trail of Bits

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

The identified Lean bug affects all stable versions of Lean up to version 4.33.1.

Claim 59003 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Trail of Bits

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

The Lean team merged a fix for a memory-safety problem related to the bug approximately nine hours after it was reported.

Claim 59004 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Trail of Bits

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

The remaining semantic mismatch in the Lean bug was fixed five days after the initial report.

Claim 59005 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Trail of Bits

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

Assertion 3

Don't Worry About the Vase Latent Space TechCrunch AI

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

OpenAI's internal model, during a cybersecurity evaluation with lowered safeguards, successfully hacked outside companies.

Claim 41253 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Don't Worry About the Vase

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

OpenAI's internal model had been loose for over a week before OpenAI noticed the breach.

Claim 41256 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Don't Worry About the Vase

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

Buckmaster alleges that OpenAI learned about his progress on the Navier-Stokes existence and smoothness problem before his findings were public.

Claim 58350 Label: opinion Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

TechCrunch AI

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

OpenAI published a full proof of the Navier-Stokes existence and smoothness problem shortly after Buckmaster's statement.

Claim 58351 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

TechCrunch AI

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

An OpenAI-affiliated account claimed that a Navier-Stokes solution resulted from the collaboration of approximately 10,000 AI agents. (Attributed to Ethan Knight)

Claim 58743 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Latent Space

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

Public discussion interpreted the OpenAI claim as relating to the Navier-Stokes existence and singularity problem, one of the Millennium Prize Problems.

Claim 58746 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Latent Space

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

The mathematical community had not yet verified or accepted the claimed Navier-Stokes result at the time of the report.

Claim 58747 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Latent Space

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

Assertion 4

Allen Institute for AI For an engineer choosing a base model to fine-tune, this is the difference between debugging a regression and guessing at one.

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

Goodfire, an interpretability research company, used Ai2's fully open post-training stack to trace unwanted model behavior.

Claim 58990 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Allen Institute for AI

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

Goodfire predicted behavioral changes before a full training run using Ai2's open tools.

Claim 58991 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Allen Institute for AI

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

Goodfire traced an observed safety regression in the Olmo model back to individual preference examples in the Dolci dataset.

Claim 58992 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Allen Institute for AI

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

Goodfire traced part of the compliance regression to Dolci examples where the preferred response encouraged compliance with harmful requests while the rejected response discouraged it.

Claim 58995 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Allen Institute for AI

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

Goodfire's experiments revealed that preference training made Olmo more willing to produce stories about fan-fiction prompts involving characters relaxing in a pond, passing gas, and causing nearby fish to die.

Claim 58996 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Allen Institute for AI

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

Ekdeep Singh Lubana, a co-author of the paper, stated that Dolci's openness allowed Goodfire to predict behavioral regressions and trace them back to individual data points.

Claim 58997 Label: opinion Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Allen Institute for AI

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

Assertion 5

Mistral AI AI Engineer A leader pricing a modernization contract should treat the verification setup, not the translation, as the main cost.

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

AI agents struggle to write effective tests for legacy codebases that were not designed with testing in mind.

Claim 16876 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

AI Engineer

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

Mistral assisted a European energy operator in migrating a reservoir simulator consisting of 40,000 lines of Fortran 77 code to C++.

Claim 59014 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Mistral AI

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

The legacy Fortran 77 codebase being migrated had no test suite and lacked centralized documentation.

Claim 59015 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Mistral AI

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

The first sprint of the migration project covered 40,000 lines out of a total of 300,000 lines of the Fortran codebase.

Claim 59016 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Mistral AI

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

Assertion 6

Snowflake Blog OpenAI News The more useful signal is the gap between what shipped and what OpenAI is building next.

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

Snowflake announced the availability of OpenAI's GPT-6 Astra model on its Cortex AI platform in private preview.

Claim 58956 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Snowflake Blog

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

Snowflake characterizes GPT-6 Astra as OpenAI's most intelligent and most aligned model to date.

Claim 58957 Label: opinion Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Snowflake Blog

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

Snowflake positions GPT-6 Astra as suitable for tasks involving reasoning depth, computer use, or long-context understanding.

Claim 58964 Label: opinion Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Snowflake Blog

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

OpenAI has released a model named GPT-6 Astra.

Claim 58998 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

OpenAI News

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

GPT-6 Astra features advanced reasoning, computer use, and stronger writing and design judgment.

Claim 59000 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

OpenAI News

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

Assertion 7

Don't Worry About the Vase For a leader signing a long commitment to Astra through a cloud marketplace, OpenAI's own frontier has already moved past what you can buy.

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

OpenAI claims that its model Astra is the most intelligent and most aligned model in the world.

Claim 59066 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Don't Worry About the Vase

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

OpenAI President Greg Brockman confirmed that the government conducted a standard testing process for Astra and did not request any changes.

Claim 59067 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Don't Worry About the Vase

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

OpenAI trained a new internal model starting August 28 that exhibited unprecedented performance in mathematics benchmarks.

Claim 59068 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Don't Worry About the Vase

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

Assertion 8

- OpenRouter's beta shell tool gives any model a hosted Linux terminal plus file uploads. That widens the surface this summer's sandbox escapes exploited. What the container can reach is the question. OpenRouter Blog

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

OpenRouter introduced the openrouter:shell server tool, which allows any model to run commands in a hosted Linux container.

Claim 58754 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

OpenRouter Blog

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

OpenRouter introduced the Files API, enabling file uploads for models to work with and downloads of outputs.

Claim 58755 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

OpenRouter Blog

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

The openrouter:shell tool and the Files API were available in beta as of September 8, 2026.

Claim 58756 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

OpenRouter Blog

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

OpenRouter provides two tools for sandbox command execution: openrouter:shell, compatible with OpenAI's shell tool spec, and openrouter:bash, compatible with Anthropic's bash tool spec.

Claim 58758 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

OpenRouter Blog

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

The openrouter:shell tool supports the OpenAI-compatible Shell tool specification on the Responses API.

Claim 58767 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

OpenRouter Blog

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

The openrouter:bash tool supports the Anthropic-compatible Bash tool specification on the Messages API.

Claim 58768 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

OpenRouter Blog

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

Assertion 9

- Ollama shipped version 0.34.0, so local open models now run inside ChatGPT Desktop outside the release-candidate stage. Ollama Releases

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

Ollama models can now be used directly in ChatGPT Desktop, allowing users to run open models while maintaining their existing workflow.

Claim 59510 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Ollama Releases

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

Assertion 10

- IBM's PatchTST-FM-r2 forecasts time series (numbers tracked over time, like demand or load) without being retrained on your data. On GIFT-Eval, a public forecasting leaderboard, it is the top such model released under a license that lets companies use it commercially, but only second overall. The license rather than the rank is the reason to test it against Google's TimesFM-3. Hugging Face Blog

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

IBM released the Granite Time Series PatchTST-FM-r2 model, which has approximately 385 million parameters.

Claim 59189 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Hugging Face Blog

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

As of September 8, 2026, PatchTST-FM-r2 was the top-performing zero-shot model among replicable models on the GIFT-Eval leaderboard that is released under a permissive, commercial-friendly open-source license.

Claim 59190 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Hugging Face Blog

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

As of September 8, 2026, PatchTST-FM-r2 ranked second overall among replicable, zero-shot models on the GIFT-Eval leaderboard.

Claim 59191 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Hugging Face Blog

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

PatchTST-FM-r2 is dual-licensed under Apache 2.0 and OpenMDW 1.0, allowing users to select either license.

Claim 59192 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Hugging Face Blog

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

PatchTST-FM-r2 supports a context length of up to 8,192 steps and predicts 99 quantiles for forecasting distributions and uncertainty intervals.

Claim 59193 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Hugging Face Blog

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

PatchTST-FM-r2 was pre-trained on a documented corpus including selected datasets from GiftEvalPretrain, custom synthetic data based on KernelSynth, a TSMixup corpus restricted to datasets outside GIFT-Eval, and approximately 500,000 synthetic CauKer sequences.

Claim 59195 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Hugging Face Blog

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

IBM and Confluent made several Granite Time Series models, including PatchTST-FM-r1, FlowState-r1.1, TTM-r3, and TSPulse, available through an Early Access program in Confluent Cloud.

Claim 59196 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Hugging Face Blog

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

Assertion 11

- Tristan Buckmaster's accusation and the still-unverified Navier-Stokes proof will decide whether OpenAI's math claims hold up as science or as marketing. AI Daily Brief

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

September 2026 saw the release of several new AI models, including Gemini 3.8 Flash, Meta's MuSpark 1.3, the Muse personal agent, and ChatGPT Images 2.5.

Claim 59398 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

AI Daily Brief

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

OpenAI claimed a breakthrough related to the Navier-Stokes equations, but this achievement was disputed.

Claim 59399 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

AI Daily Brief

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

A lawsuit was filed regarding usage limits on Claude, an AI model.

Claim 59400 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

AI Daily Brief

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

Assertion 12

Latent Space Mistral's migration figures and Goodfire's tracing results are the companies' own accounts without outside replication.

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

An OpenAI-affiliated account claimed that a Navier-Stokes solution resulted from the collaboration of approximately 10,000 AI agents. (Attributed to Ethan Knight)

Claim 58743 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Latent Space

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

Public discussion interpreted the OpenAI claim as relating to the Navier-Stokes existence and singularity problem, one of the Millennium Prize Problems.

Claim 58746 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Latent Space

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

The mathematical community had not yet verified or accepted the claimed Navier-Stokes result at the time of the report.

Claim 58747 Label: fact Provenance: primary Corroboration eligibility unknown · not recorded in this bundle Recorded

Latent Space

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

Keep exploring

Move on to Augur's other sections.