AIRankBankEvery launch is a claim. Every claim gets a verdict.
AI Launch · Bank records

Anthropic — delivery record

Anthropic currently has 45 claims on the record at AI Rank Bank. Of the 36 that have come due and been graded, 35 shipped and 1 busted, for a delivery rate of 97.2%. 1 more ended unproven: the date passed without public evidence either way. 8 markets are still open, and 105 reader calls are riding on this company's claims.

A delivery rate is not an opinion. Every claim below was pulled from the company's own announcement coverage, given a check date when the market opened, and stamped against public evidence when the date arrived. Announcements are cheap; this page is what they cost. Companies that ship what they announce watch the number climb. Companies that announce first and quietly narrow later watch it fall — in public, one dated receipt at a time.

Claims on record45
Delivered35
Busted1
Unproven1
Still open8
Delivery rate97.2%

Every claim, on the record

The standard gets a public spec update or a second adopting organisation.
settles 2026-11-27 · check date 2026-11-27 · readers 100% deliver · 3 calls
Barclays publicly expands Claude beyond the initial rollout.
settles 2026-12-01 · check date 2026-12-01 · readers 50% deliver · 4 calls
The enterprise frontier safeguards actually ship to customers.
settles 2026-12-01 · check date 2026-12-01 · readers 75% deliver · 4 calls
Frontier Academy opens applications or names its first cohort.
settles 2026-12-02 · check date 2026-12-02 · readers 67% deliver · 3 calls
The Enterprise Frontier Safeguards system has begun rolling out to enterprise customers.
settles 2026-12-06 · check date 2026-12-06 · readers 100% deliver · 1 calls
Embedded evaluation shows up in a shipped model release (system card or docs).
settles 2026-12-18 · check date 2026-12-18 · readers 100% deliver · 1 calls
An independent lab publishes validation or follow-up work on the enzyme system.
settles 2027-01-23 · check date 2027-01-23 · readers 33% deliver · 3 calls
Anthropic rolls out the new monthly API credit to Max and Team subscribers.
✓ DELIVERED · check date 2026-10-14 · readers 0% deliver · 1 calls
Claude Docs, Slides, and Design turn on by default on Enterprise plans on October 15, 2026.
✓ DELIVERED · check date 2026-10-15 · readers 50% deliver · 2 calls
OSS Scanner is open for enrollment, offering critical open-source projects free periodic security scans.
✓ DELIVERED · check date 2026-10-16 · readers 80% deliver · 5 calls
Anthropic publishes the evaluation results this work commits to.
✗ BUSTED · check date 2026-10-31 · readers 0% deliver · 1 calls
Anthropic publishes the program's criteria or its first verified participants.
✓ DELIVERED · check date 2026-11-06 · readers 50% deliver · 6 calls
Claude Sonnet 5 remains the default model for Free and Pro plans.
✓ DELIVERED · check date 2026-11-06 · readers 0% deliver · 1 calls
Claude Opus 5 remains the default model on Claude Max.
✓ DELIVERED · check date 2026-11-06 · readers 0% deliver · 1 calls
Recipients of Anthropic's $5M wellbeing-evaluation grants have been publicly announced.
? UNPROVEN · check date 2026-11-06 · readers 100% deliver · 1 calls
The Anthropic Economic Index connector is available to Claude users in chat.
✓ DELIVERED · check date 2026-11-06 · readers 0% deliver · 1 calls
The 10,000 Claude seats for scientists are open for sign-up.
✓ DELIVERED · check date 2026-11-06 · readers 0% deliver · 1 calls
Claude Opus 5.5 is available on AWS, Google Cloud and Azure.
✓ DELIVERED · check date 2026-11-06 · readers 100% deliver · 3 calls
Claude Haiku 5.5 has been released.
✓ DELIVERED · check date 2026-11-06 · readers 50% deliver · 2 calls
Anthropic's updated Usage Policy takes effect on November 12, 2026.
✓ DELIVERED · check date 2026-11-12 · readers 86% deliver · 7 calls
The program names its first verified organisations or exits beta.
✓ DELIVERED · check date 2026-11-17 · readers 75% deliver · 4 calls

Other companies in the bank

OpenAI · Google DeepMind · Meta AI · DeepSeek · xAI · Mistral AI · Microsoft · NVIDIA

How claims are judged: the sentence itself is the test, the publisher's page is the source of record, and verdicts — delivered, busted, unproven — are permanent. Readers price every claim with points on the homepage and build their own records on the leaderboard.