Zero Anthropic staff engagement on 13 documented model failures from a paying customer

Summary

I have 13 open issues on this repository documenting reproducible model failures in Claude Code. Several include forensic audits, cost analyses, session transcripts, and contributed fixes. Anthropic staff have not responded to any of them.

The issues

| # | Title | Filed | Comments | Anthropic response |
|---|-------|-------|----------|-------------------|
| #28469 | Opus 4.6 comprehensive regression | 2026-02 | 22 | None |
| #59515 | SessionStart hooks lack hard enforcement | 2026-05 | 2 | None |
| #59529 | Memory directives loaded but not honoured | 2026-05 | 7 | None |
| #61107 | Opus 4.7 silently discards user input | 2026-05 | 10 | None |
| #61167 | Opus 4.7 fabricates agent dispatches | 2026-05 | 14 | None |
| #62571 | Model-version pinning request | 2026-05 | 2 | None |
| #66049 | Opus refuses to use available tools | 2026-06 | 0 | None |
| #66050 | Refuses to follow explicit instructions | 2026-06 | 0 | None |
| #66054 | Claims work is done without verifying | 2026-06 | 3 | None |
| #66143 | Forgets known facts despite memory system | 2026-06 | 3 | None |
| #72651 | Model ignores CLAUDE.md instructions | 2026-07 | 1 | None |
| #72655 | Contributed fix: enforcement hooks | 2026-07 | 1 | None |
| #72657 | Cost of non-compliance documentation | 2026-07 | 0 | None |

What I have provided

• Forensic audits with session transcripts, curl logs, and ground-truth comparisons (#61167: 34% phantom dispatch rate across 5 sessions)
• Cost documentation showing hours wasted per failure pattern (#59529, #72657)
• Reproducible steps with specific model versions, OS, and configuration
• A contributed fix with installable enforcement hooks (#72655) that other users can deploy today
• Community validation from other paying users confirming the same patterns (#28469, #59515, #61167)

What Anthropic has provided

Nothing. No acknowledgment, no triage, no "we're looking into it," no timeline. The only Anthropic-affiliated activity on these threads is the auto-close bot trying to mark them as duplicates.

The problem

I pay for Claude Max. I use Claude Code daily for production work (healthcare startup). When the model fabricates agent dispatches, ignores explicit instructions, or claims work is done without checking, it costs real time and real money.

I have documented these failures thoroughly, contributed a fix, and engaged constructively with the community. The response from Anthropic is silence.

If there is a better channel for production-impacting bugs from paying customers, say so. If these issues are deprioritized, say that too. Silence is not acceptable when people are paying for the product and reporting reproducible failures.

— Nofyah