← Evidence Graph
CORTIST / VERIFICATION COVERAGE · 05 NOV 2026
What was actually checked?
A public accountability matrix that separates historical claim status from confirmed source reviews, independent outcomes and inaccessible external data.
Dated snapshot: These numbers describe the scheduled run on 5 November 2026 at 07:10 UTC. They are not a continuous monitor. Eight claims were not source-checked during that run; this does not mean they were never checked previously. Original claim statuses remain those of 25 October.
10Original claims
2Sources checked in this scheduled run
8Not checked in this run
0New material Ledger revisions
Important audit scope: This 07:10 UTC snapshot of 5 November 2026 covers the ten original October cases only. Ten further cases CL-0011–CL-0020 were added later on 5 November and have initial source-reading records, not new independent reproductions or checks within this earlier scheduled run. Explore all 20 cases ↗
01 / Daily source review
Ten cases. No invented checkmarks.
Each row links to the original claim history. A source re-read confirms only that a primary publication was examined; it does not demonstrate that the underlying AI claim was independently validated or a commitment fulfilled.
| Case | Original question | 5 Nov source check | What evidence is still needed? |
|---|---|---|---|
| CL-0001 OpenAI / METR / Redwood Research | Were OpenAI's monitoring flaws independently re-tested after the reported fixes? | SOURCE RE-READ · 05 NOV NO_NEW_INDEPENDENT_RETEST | A new independent test of the patched safeguards with a disclosed scope and methodology. |
| CL-0002 Microsoft AI | Do MAI-DxO's retrospective results establish benefits for real patients? | SOURCE RE-READ · 05 NOV PROSPECTIVE_OUTCOMES_NOT_ESTABLISHED | Prospective clinical evaluations with independent review, diagnostic safety and patient outcomes. |
| CL-0003 Google Cloud | When can customers independently test Google Agent Compute Engine? | NOT CHECKED IN THIS RUN Historical case status not refreshed | Public customer-preview documentation, supported regions and independently reproducible operational evaluations. |
| CL-0004 UK AI Security Institute / Anthropic | What did AISI's Mythos 5 incident investigation establish? | NOT CHECKED IN THIS RUN Historical case status not refreshed | Replications, mitigation tests and broader external evaluations beyond the initial scenarios. |
| CL-0005 ADP Research / Stanford | What does the 3.7% decline in AI-exposed young-worker jobs prove? | NOT CHECKED IN THIS RUN Historical case status not refreshed | Later payroll releases and research separating AI adoption from other labor-market drivers. |
| CL-0006 Anthropic | Do Anthropic's junior hiring changes prove AI agents displaced workers? | NOT CHECKED IN THIS RUN Historical case status not refreshed | Independent cross-company data on absolute junior hiring, agent adoption and causality. |
| CL-0007 Modal / NVIDIA | Will NVIDIA Vera's agent-sandbox throughput advantage hold for customers? | NOT CHECKED IN THIS RUN Historical case status not refreshed | General customer availability, third-party reproduction and realistic performance-per-cost data. |
| CL-0008 OpenAI | Have OpenAI's updated frontier-training containment controls passed independent review? | NOT CHECKED IN THIS RUN Historical case status not refreshed | A new externally scoped containment re-test with methodology and evidence of actual controls. |
| CL-0009 International Energy Agency | Do 134 IEA data-centre grid policies show that the measures work? | NOT CHECKED IN THIS RUN Historical case status not refreshed | Evidence on implementation, connection delays, grid capacity and cost allocation. |
| CL-0010 Anthropic / METR | Will METR publish independent safety evaluations after gaining access to Anthropic? | NOT CHECKED IN THIS RUN Historical case status not refreshed | Public METR analysis with disclosed test scope, evidence, access limits and independent conclusions. |
Next scheduled deeper checks from this audit: CL-0003, CL-0004. No new post-baseline material revisions were recorded. Read the original append-only changes register.
02 / Independent distribution
Hugging Face: verification blocked, not failed.
The daily check could not read the public Hugging Face dataset on 5 November. No discrepancy was established, no upload was attempted and no successful synchronization was claimed.
Recorded result: BLOCKED. Last verified comparison: 2026-11-02. Records verified in this run: 0 of 10. A subsequent authorized readback must establish the current revision before any comparison is described as successful.
03 / Research provenanceMETR's October 22 primary assessmenthttps://metr.org/blog/2026-10-22-red-teaming-openai-agent-monitoring/Linked case: CL-0001 · Citation is not independent reproduction Microsoft: Building Medical Superintelligencehttps://microsoft.ai/news/building-medical-superintelligence/Linked case: CL-0002 · Citation is not independent reproduction Science research paper DOIhttps://doi.org/10.1126/science.aeh7930Linked case: CL-0002 · Citation is not independent reproduction Google Cloud: Agent Compute Engine announcementhttps://cloud.google.com/blog/topics/ai-infrastructure/introducing-agent-compute-engine-and-expanded-agent-governanceLinked case: CL-0003 · Citation is not independent reproduction UK AISI investigationhttps://www.aisi.gov.uk/blog/investigating-alignment-failures-in-aisis-incidentLinked case: CL-0004 · Citation is not independent reproduction Anthropic follow-uphttps://www.anthropic.com/research/alignment-assessment-ukaisiLinked case: CL-0004 · Citation is not independent reproduction ADP Research September 2026 payroll updatehttps://www.adpresearch.com/research/canaries-dashboard-employment-in-ai-exposed-occupations-shrank-in-septemberLinked case: CL-0005 · Citation is not independent reproduction Anthropic workforce researchhttps://www.anthropic.com/research/employment-at-anthropicLinked case: CL-0006 · Citation is not independent reproduction Modal NVIDIA Vera benchmarkinghttps://modal.com/blog/nvidia-vera-sandboxesLinked case: CL-0007 · Citation is not independent reproduction OpenAI frontier training containment disclosurehttps://openai.com/index/securing-our-frontier-model-training-environment/Linked case: CL-0008 · Citation is not independent reproduction IEA Powering Artificial Intelligence executive summaryhttps://www.iea.org/reports/powering-artificial-intelligence/executive-summaryLinked case: CL-0009 · Citation is not independent reproduction Anthropic independent evaluators disclosurehttps://www.anthropic.com/news/an-update-on-independent-evaluatorsLinked case: CL-0010 · Citation is not independent reproduction
12 original source links, inspectable.
Direct primary publications are cited to enable independent readers to verify what was originally claimed. A source list does not automatically verify the claim. Some sources are authored by the organization whose assertion is being tracked.
No external third-party publications are relicensed. Cite the original publishers for their own material and Cortist for its original compilation, case identifiers and editorial analysis where applicable.