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.
On September 18 it added an outside tester: Accenture will run embedded evaluation of Anthropic's frontier models, its most capable ones, through Faculty, Accenture's specialist AI unit, and each company intends to put at least $1 billion into evaluation capacity over five years Anthropic News.
Assertion status: No spot-check verdict is published for this assertion.
Anthropic has entered into a partnership with Accenture to conduct independent embedded evaluation of its frontier AI models.
Claim 63612Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 2
Anthropic has also promised the independent evaluators METR and Redwood Research a level of access it calls unprecedented, and OpenAI has pledged to follow with embedded investigators of its own TechCrunch AI Don't Worry About the Vase.
Assertion status: No spot-check verdict is published for this assertion.
Anthropic CEO Dario Amodei published an essay proposing the embedding of third-party evaluators inside frontier AI companies to report safety incidents and assess alignment. (Published 2026-09-16)
Claim 62515Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 3
Accenture is an anchor partner in the Claude Partner Network, the $100 million program in which Anthropic enlisted consultancies to work with its models, and this month it formed a joint unit with Google Cloud to deploy engineers for Google's AI tools Nate Jones TechCrunch AI.
Assertion status: No spot-check verdict is published for this assertion.
Anthropic committed $100 million to the Claude Partner Network and engaged Accenture, Deloitte, Cognizant, and Infosys as anchor partners.
Claim 14928Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 4
The yardstick was set a week earlier: Dario Amodei's pacing plan described embedding evaluators from organizations like METR with badges, desks, and access comparable to internal risk teams, and the first signed partner under that banner is a firm that sells the product TechCrunch AI.
Assertion status: No spot-check verdict is published for this assertion.
Anthropic plans to embed third-party evaluators from organizations like METR inside AI companies to verify pacing and safety commitments and to ensure reporting of safety incidents.
Claim 61214Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
These embedded evaluators will have company badges, desks, laptops, and access comparable to internal risk teams, except where law or contracts restrict access, according to Anthropic CEO Dario Amodei.
Claim 61215Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 5
The contrast with the independent track is the point: as of July, Anthropic was still in dialogue with METR over a review with access to all transcripts Anthropic News.
Assertion status: No spot-check verdict is published for this assertion.
Anthropic is in dialogue with METR, an independent AI evaluation organization, to conduct a third-party review including access to all transcripts and sampling access to the relevant models.
Claim 55277Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 6
OpenRouter's new tutorial makes a point every agent builder should absorb: an agent, a program that carries out multi-step tasks on its own, can pass every scripted test and still give a bad answer, such as calling the right tool and then leaving the policy details out of its reply, which is why the guide has a second model score outputs against written criteria OpenRouter Blog.
Assertion status: No spot-check verdict is published for this assertion.
OpenRouter published a tutorial titled "LLM-as-a-Judge: Score AI Agent Outputs Automatically" on September 14, 2026.
Claim 63366Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
The source asserts that an agent can pass all deterministic tests yet still produce a poor answer, such as correctly calling a tool but failing to include relevant policy details in the final response.
Claim 63367Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
The source defines LLM-as-a-judge evaluation as a method where a second model reviews an agent's output against written criteria to provide a repeatable score for open-ended responses.
Claim 63368Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 7
AWS now sells AgentCore Evaluations, which scores live agent conversations the same way, and developers are using TypeSafe's Jev, a model that only classifies and routes and which its users say cannot write free text, as a cheap judge AWS Machine Learning Blog Latent Space Latent Space.
Assertion status: No spot-check verdict is published for this assertion.
AgentCore Evaluations is a quality assessment framework that continuously scores live agent interactions using an LLM-as-a-Judge methodology.
Claim 60830Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Community analysis indicates that Jev cannot produce free-form text and functions as a structured classifier or routing policy rather than a general language model.
Claim 62210Label: opinionProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Anthropic released a 'Projects' feature in Claude Code that allows a single conversation to spawn parallel cloud sessions and pass context between them, with internal staff describing it as a higher-level coordinator abstraction.
Claim 63636Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Google updated Gemini managed agents with an Antigravity-based harness, a Credentials API, and a Files API, claiming up to 30% lower costs and 22% higher cache hits.
Claim 63637Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Developers are treating TypeSafe's Jev as a routing and judgment primitive rather than a chatbot, using it for LLM-as-judge, harness routing, and structured outputs.
Claim 63638Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 8
The catch has not moved: these judges need bias calibration, and practitioners describe them as nondeterministic, scoring the same input differently on different runs Together AI Blog AI Engineer.
Assertion status: No spot-check verdict is published for this assertion.
LLM-as-a-judge evaluators are described as nondeterministic, producing different answers for the same input on different runs. fact
Claim 28386Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Using LLMs to judge other LLMs (LLM-as-a-judge) is a scalable evaluation strategy allowing customized criteria but requires bias calibration and careful interpretation.
Claim 32119Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Trail of Bits asserts that 1Password's reported 26% clean-fix rate is misleading because it includes experiments where agents were instructed to apply incorrect fixes.
Claim 61657Label: opinionProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Trail of Bits asserts that 1Password's reported 26% clean-fix rate is misleading because it combines trials where agents could test their patches with trials where testing was prohibited.
Claim 61658Label: opinionProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Trail of Bits asserts that 1Password's reported 26% clean-fix rate is misleading because the study selected vulnerabilities specifically for their complexity rather than as a representative sample.
Claim 61659Label: opinionProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Trail of Bits asserts that 1Password's reported 26% clean-fix rate is misleading because the models were run at default reasoning settings rather than their highest available settings.
Claim 61660Label: opinionProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Trail of Bits states that in trials from 1Password's data where agents could test patches and were not instructed to apply wrong fixes, 86% of the generated patches blocked the supplied exploit.
Claim 61661Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 10
Before it looked at a line of the Miden VM, the software it was preparing to audit, the firm put half a year into letting agents write its instruments, among them a decompiler and a static analysis engine (programs that read compiled code and scan it for flaws); the audit then turned up an unchecked input that could let an attacker forge signatures and drain account holders' funds, and it closed with 95 machine-checked proofs in Lean, a system that verifies math by computer Trail of Bits.
Assertion status: No spot-check verdict is published for this assertion.
Trail of Bits spent six months using AI agents to build custom tooling, including an LSP server, a decompiler, a static analysis engine, and a Lean model, to prepare for an audit of the Miden VM.
Claim 63662Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Trail of Bits identified an unvalidated prover-supplied input in the Miden VM that could allow a malicious prover to forge Falcon signatures and steal funds from Miden account holders.
Claim 63663Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 12
Les Karpas, who leads physical AI at Nvidia's startup program Inception, says it has not arrived, and his reason is structural: language models had an internet's worth of text to learn from, and no equivalent dataset exists for how bodies move through the world TechCrunch AI.
Assertion status: No spot-check verdict is published for this assertion.
Les Karpas, Nvidia Inception’s Global Head of Physical AI, stated at TechCrunch Disrupt 2026 that robotics has yet to experience a breakthrough moment comparable to ChatGPT's impact on AI. The source attributes this claim to Karpas. fact null null
Claim 63410Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Les Karpas identified the absence of an internet-wide dataset for physical AI as the core limitation facing general-purpose robots, noting that language AI had such datasets while robotics lacks them. The source attributes this explanation to Karpas. fact null null
Claim 63411Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
TechCrunch Disrupt 2026 will take place from October 13-15 at San Francisco’s Moscone West, featuring a session by Les Karpas on the Real World AI Stage regarding physical AI bottlenecks. The source states this as a scheduled event. fact 2026-10-13T00:00:00+00:00 null null
Claim 63414Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Les Karpas will discuss the limitations and opportunities in AI and robotics during his session at TechCrunch Disrupt 2026. The source reports this as an upcoming event detail. fact 2026-09-16T15:00:00+00:00 null null
Claim 63416Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 13
- Anthropic's next system card, the report a lab publishes alongside a model, will show whether Faculty's evaluators appear as contributors or stay off the page. Anthropic News
Assertion status: No spot-check verdict is published for this assertion.
Anthropic has entered into a partnership with Accenture to conduct independent embedded evaluation of its frontier AI models.
Claim 63612Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 14
- OpenAI has pledged embedded investigators without naming them; the name will reveal whether it follows the independent-evaluator or the consultancy model. Don't Worry About the Vase
Assertion status: No spot-check verdict is published for this assertion.
OpenAI pledged to join Anthropic in taking the step of deploying embedded investigators for AI safety.
Claim 63018Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 15
- METR and Redwood Research's promised access should surface as a published review; whether it lands before or after Accenture's first output is the tell. TechCrunch AI
Assertion status: No spot-check verdict is published for this assertion.
Anthropic CEO Dario Amodei published an essay proposing the embedding of third-party evaluators inside frontier AI companies to report safety incidents and assess alignment. (Published 2026-09-16)
Claim 62515Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 16
- Google claims its updated Gemini managed agents cut costs by up to 30%, a figure worth checking against your own bills before you migrate. Latent Space
Assertion status: No spot-check verdict is published for this assertion.
Anthropic released a 'Projects' feature in Claude Code that allows a single conversation to spawn parallel cloud sessions and pass context between them, with internal staff describing it as a higher-level coordinator abstraction.
Claim 63636Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Google updated Gemini managed agents with an Antigravity-based harness, a Credentials API, and a Files API, claiming up to 30% lower costs and 22% higher cache hits.
Claim 63637Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Developers are treating TypeSafe's Jev as a routing and judgment primitive rather than a chatbot, using it for LLM-as-judge, harness routing, and structured outputs.
Claim 63638Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 17
- Les Karpas speaks at TechCrunch Disrupt on October 13-15; listen for any dataset effort Nvidia names rather than a timeline. TechCrunch AI
Assertion status: No spot-check verdict is published for this assertion.
Les Karpas, Nvidia Inception’s Global Head of Physical AI, stated at TechCrunch Disrupt 2026 that robotics has yet to experience a breakthrough moment comparable to ChatGPT's impact on AI. The source attributes this claim to Karpas. fact null null
Claim 63410Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Les Karpas identified the absence of an internet-wide dataset for physical AI as the core limitation facing general-purpose robots, noting that language AI had such datasets while robotics lacks them. The source attributes this explanation to Karpas. fact null null
Claim 63411Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
TechCrunch Disrupt 2026 will take place from October 13-15 at San Francisco’s Moscone West, featuring a session by Les Karpas on the Real World AI Stage regarding physical AI bottlenecks. The source states this as a scheduled event. fact 2026-10-13T00:00:00+00:00 null null
Claim 63414Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Les Karpas will discuss the limitations and opportunities in AI and robotics during his session at TechCrunch Disrupt 2026. The source reports this as an upcoming event detail. fact 2026-09-16T15:00:00+00:00 null null
Claim 63416Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 18
The $1 billion figures are stated intent over five years, not committed spending, and the announcement we drew on describes capacity rather than any evaluation output Anthropic News.
Assertion status: No spot-check verdict is published for this assertion.
Anthropic has entered into a partnership with Accenture to conduct independent embedded evaluation of its frontier AI models.
Claim 63612Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Assertion 19
Anthropic's promised access for METR and Redwood Research remains a stated commitment in our evidence, with the July METR review described as a dialogue rather than a result, and OpenAI's pledge as reported names no evaluator TechCrunch AI Anthropic News.
Assertion status: No spot-check verdict is published for this assertion.
Anthropic is in dialogue with METR, an independent AI evaluation organization, to conduct a third-party review including access to all transcripts and sampling access to the relevant models.
Claim 55277Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded
No stored spot-check names this claim in this edition.
Anthropic CEO Dario Amodei published an essay proposing the embedding of third-party evaluators inside frontier AI companies to report safety incidents and assess alignment. (Published 2026-09-16)
Claim 62515Label: factProvenance: primaryCorroboration eligibility unknown · not recorded in this bundleRecorded