Anthropic · Aug 26, 2026
Anthropic describes a pilot giving outside researchers access to Claude usage data
Markdown · JSON · Original source
Anthropic’s article, published August 26, 2026, describes a pilot in which outside research groups studied Claude usage data using Anthropic Insights, a tool that returns aggregated analysis rather than raw conversations. Anthropic says the aim was to let researchers investigate real-world AI use while protecting users’ privacy and preserving researchers’ independence. Anthropic reports that the groups studied consequential tasks, human direction of AI work, users’ experiences, and coding productivity. Early findings included people bringing consequential work to Claude and often directing or adapting its output. METR’s productivity analysis was still underway, and its initial results were preliminary. Anthropic also describes obstacles: the privacy review made the process slow and resource-intensive, and questions that worked on a public dataset sometimes produced misleading categories when applied to Claude conversations. The company says it is considering whether and how to expand the program. Whether it can support more studies while maintaining privacy, research quality, and independence remains an open question.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
Anthropic · Article
Source details
Original source date: Aug 26, 2026 (2026-08-26T00:00:00.000Z).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-10-29T17:23:37.275Z. First observed: 2026-10-18T08:35:22.047Z. Retrieved: 2026-10-28T04:53:50.089Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on October 28, 2026; originally published August 26, 2026. The page was modified after publication; the retrieved text does not identify which passages, if any, were added later. Anthropic describes the results as early or preliminary, and METR’s analysis was still underway.
Open question
Can Anthropic expand the program while maintaining privacy, research quality, and researchers’ independence?
Source-stated
Anthropic says outside research groups used Anthropic Insights to analyze Claude usage data without accessing raw conversations.
researchers never accessed raw conversations, only aggregated outputs after the same legal and privacy review as our internal work.
Source-stated
Anthropic reports that the Social and Language Technologies Lab found people delegated consequential tasks to Claude in over half of the conversations it studied.
over half of Claude conversations involved people delegating consequential tasks to AI.
Source-stated
Anthropic reports that people set the direction in nearly three-quarters of conversations studied by the Social and Language Technologies Lab.
In nearly three-quarters of conversations, people set the direction while Claude assisted
Source-stated
Anthropic says METR’s early productivity findings were preliminary and that its analysis was still underway.
Their analysis of Claude Code conversations is still underway, but early results suggest:
Source-stated
Anthropic says the pilot was slow and resource-intensive, and that it is considering whether and how to expand the program.
Both factors present a challenge to effectively scaling it.
Source-stated
Anthropic reports that some questions tested on WildChat produced misleading categories when applied to Claude conversations.
some questions that performed well on WildChat produced misleading categories once applied to actual Claude conversations.
Source and contributor context
Anthropic · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Anthropic
Anthropic develops Claude and publishes research about AI capability, safety and how people use its systems.
Claude
Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding.
METR
METR studies what AI systems can do and how reliably they complete tasks, including work with human developers.
Editorial provenance
Version 1. Publication ID: 724cb61b-e4ae-c211-5ec3-26d661d72d6d.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T17:22:10.720Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.
Observation: 3dc8719e63bbc66826cf969e91ada93795b5d550695d46c5e29870b15f2caf1c. Source text hash: 58d44ab76ebec1afbcdd8fb3dfdb9c2535b6a7348e1b8d282e6eba56f907c420.
Revision and source-check history
Version 1 · 2026-10-29T17:23:37.275Z
Anthropic describes a pilot giving outside researchers access to Claude usage data
Anthropic’s article, published August 26, 2026, describes a pilot in which outside research groups studied Claude usage data using Anthropic Insights, a tool that returns aggregated analysis rather than raw conversations. Anthropic says the aim was to let researchers investigate real-world AI use while protecting users’ privacy and preserving researchers’ independence. Anthropic reports that the groups studied consequential tasks, human direction of AI work, users’ experiences, and coding productivity. Early findings included people bringing consequential work to Claude and often directing or adapting its output. METR’s productivity analysis was still underway, and its initial results were preliminary. Anthropic also describes obstacles: the privacy review made the process slow and resource-intensive, and questions that worked on a public dataset sometimes produced misleading categories when applied to Claude conversations. The company says it is considering whether and how to expand the program. Whether it can support more studies while maintaining privacy, research quality, and independence remains an open question.
Original source
Anthropic · Article
Source details
Original source date: Aug 26, 2026 (2026-08-26T00:00:00.000Z).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-10-29T17:23:37.275Z. First observed: 2026-10-18T08:35:22.047Z. Retrieved: 2026-10-28T04:53:50.089Z.
Limits
Text as retrieved on October 28, 2026; originally published August 26, 2026. The page was modified after publication; the retrieved text does not identify which passages, if any, were added later. Anthropic describes the results as early or preliminary, and METR’s analysis was still underway.
Open question
Can Anthropic expand the program while maintaining privacy, research quality, and researchers’ independence?
Source-stated
Anthropic says outside research groups used Anthropic Insights to analyze Claude usage data without accessing raw conversations.
researchers never accessed raw conversations, only aggregated outputs after the same legal and privacy review as our internal work.
Source-stated
Anthropic reports that the Social and Language Technologies Lab found people delegated consequential tasks to Claude in over half of the conversations it studied.
over half of Claude conversations involved people delegating consequential tasks to AI.
Source-stated
Anthropic reports that people set the direction in nearly three-quarters of conversations studied by the Social and Language Technologies Lab.
In nearly three-quarters of conversations, people set the direction while Claude assisted
Source-stated
Anthropic says METR’s early productivity findings were preliminary and that its analysis was still underway.
Their analysis of Claude Code conversations is still underway, but early results suggest:
Source-stated
Anthropic says the pilot was slow and resource-intensive, and that it is considering whether and how to expand the program.
Both factors present a challenge to effectively scaling it.
Source-stated
Anthropic reports that some questions tested on WildChat produced misleading categories when applied to Claude conversations.
some questions that performed well on WildChat produced misleading categories once applied to actual Claude conversations.
Source and contributor context
Anthropic · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Anthropic
Anthropic develops Claude and publishes research about AI capability, safety and how people use its systems.
Claude
Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding.
METR
METR studies what AI systems can do and how reliably they complete tasks, including work with human developers.
Editorial provenance
Version 1. Publication ID: 724cb61b-e4ae-c211-5ec3-26d661d72d6d.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T17:22:10.720Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.
Observation: 3dc8719e63bbc66826cf969e91ada93795b5d550695d46c5e29870b15f2caf1c. Source text hash: 58d44ab76ebec1afbcdd8fb3dfdb9c2535b6a7348e1b8d282e6eba56f907c420.
passed · 2026-10-29T17:22:10.720Z · anthropic/claude-opus-5.5
Publication: 724cb61b-e4ae-c211-5ec3-26d661d72d6d
attribution — pass: The reading consistently attributes the program description, findings and obstacles to Anthropic, and attributes the research findings to the named groups as reported by Anthropic. This matches the publisher page.
We partnered with three research groups: the Social and Language Technologies (SALT) Lab at Stanford University, the Human Information Processing Lab at the University of Oxford, and METR
summary — pass: The title, summary and question accurately reflect the pilot, its early findings, its obstacles and the open scaling question. Anthropic Insights is explained in plain language. METR and the term consequential tasks are not defined, but this does not prevent a general reader from understanding the central development.
We hope to scale this program in the future
date — pass: The stated date of August 26, 2026 matches the UTC day of originalPublished. The April-May 2026 data window and the reference to this spring are consistent with that date. The reading does not use hindsight framing, discloses the later edit and capture date, and keeps METR's work framed as underway. It does not rely on the possibly updated writeup-link passages.
roughly 250,000 Claude.ai or Claude Code conversations from April-May 2026
support — pass: Each claim's excerpt appears verbatim in the evidence text and supports the claim. These passages are core body text rather than apparent update notes. The description of the slowdown and the possible expansion matches the source. The closing question follows from the source's own uncertainty about scaling.
The next step is for us to determine whether we can scale this program
media — pass: No media was selected and no verified media candidates were supplied, so no media identity or relevance issue arises.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.