Evidence Boundary
- These are public customer reports filed in an Anthropic-owned repository, not company-authored findings.
- Several issues were marked as duplicates, which supports clustering rather than nine independent root causes.
- Issue and comment counts are not a unique-user count.
- The issue bodies establish access, entitlement, and usage-credit-gate symptoms; they generally do not prove charges, refunds, or a combined loss.
- Feedback IDs, usernames, screenshots, and other reporter identifiers are intentionally omitted.
Issue Burst
Nine closely related issues were opened between 18:17:40 and 19:24:29 UTC on July 17.
| Issue | Opened UTC | Publicly stated context | State / evidentiary value |
|---|---|---|---|
| #78610 | 18:17:40 | Active subscription usage reportedly remained; comments describe mid-conversation interruption, 45% and 50% remaining, missing usage meter, macOS/Windows, and web/Claude Code impact. | Open; 13 comments; strongest same-minute multi-user cluster. |
| #78611 | 18:29:14 | Original report said “I have limits”; a comment said the account was not close to its limit. | Open; labeled duplicate; five comments. |
| #78614 | 18:33:50 | Usage credits were required mid-session even though included access was expected to continue. | Open; labeled duplicate and area:cost. |
| #78615 | 18:34:27 | Original report said usage was around 12%; a second user reported the same issue on Windows 11. | Closed later as a duplicate of #78614. |
| #78616 | 18:34:57 | Max plan showed 24% session and 28% weekly usage while the model required credits. | Open; labeled duplicate and area:cost. |
| #78620 | 18:44:11 | Included access was expected through July 19; comments recorded the credit gate and its apparent fix. | Open; labeled duplicate and area:cost. |
| #78623 | 18:49:16 | New Claude Code sessions requested API credits before the announced metering date. | Closed by the reporter after the transient issue appeared fixed. |
| #78633 | 19:22:03 | The reporter said the same failure appeared in Claude Code and Claude.ai while another included model still worked. | Open; labeled duplicate and area:cost; comments add Linux and continuing-impact reports. |
| #78634 | 19:24:29 | Usage credits were required two days before the expected billing change. | Open; labeled area:cost. |
What This Establishes
- An abrupt, tightly clustered July 17 event.
- Reports across macOS, Linux, Windows, Claude Code, and Claude.ai.
- Stated included-plan headroom including approximately
12%,24%session /28%weekly,45%, and50%usage positions. - Mid-session interruption and disappearance of the included-model usage meter.
- Issue deduplication around the same “usage credits required” failure.
The tracker does not establish how many affected accounts incurred credit debits, auto-reloads, invoices, or payment-method charges. Anthropic's internal records are required for that.
Relationship to the Official Incident
The issue burst overlaps Anthropic's status timeline. A public comment on #78610 records the 18:48 UTC update that Anthropic had applied a fix for the erroneous usage-credit requirement.
Questions Only Anthropic Can Resolve
- How many accounts encountered the credit gate while included entitlement remained?
- How many of those accounts had usage credits enabled?
- How many credit debits, auto-reloads, invoices, or payment-method charges occurred?
- Did Anthropic reconcile complete July 17 ledgers or only a narrow incident window?
- Why do some account corrections appear as expiring account credit instead of payment-method refunds?