Claude (Anthropic)

ai-chat · Anthropic · founded 2021

Nerf Index

on watch

Executive Summary

2 major changes on record · roughly one every 107 days.

Primary use case: General chat + reasoning.

Risk Profile

Total receipts

Critical

Major

Last 90 days

Peak nerf score

minor

Avg severity

~107d

Cadence

Series F

Funding stage

12-Month Forecast

33%

Next 30 days

56%

Next 60 days

70%

Next 90 days

Most likely change kind: feature-gated · Expected severity: minor

Observed 5 receipts over 426 days (cadence ~85d). Last receipt 20 days ago. Most common kind: feature-gated.

Receipts (chronological)

2026-06-09 · minor · nerf 18 · feature-gated
Claude Fable 5 hard-refuses cybersecurity, bio/chem, and 'distillation' prompts - silently routing them to the older Opus 4.8
Anthropic launched Claude Fable 5 on June 9, 2026 - a new 'Mythos-class' flagship 'made safe for general use' at $10/M input and $50/M output. The catch for security and science users: when Fable 5 detects a request in three domains - offensive/exploitation cybersecurity, most biology and chemistry, and prompts that try to 'distill' Claude's capabilities - it does not answer; the request is automatically handled by the older Claude Opus 4.8 instead. Anthropic says >95% of sessions hit no fallback and users are told when it happens. The unrestricted version, Claude Mythos 5, is gated to authorized cybersecurity partners and selected biology researchers only.
Predicted impact: 9-21% of paid users · ~$3-$30 per user
2026-05-14 · major · nerf 47 · billing-model-shift
Anthropic moves Claude Agent SDK / programmatic usage off subscription pools to a separately-metered API-rate credit (June 15)
Per Anthropic's own help center, starting June 15, 2026 programmatic Claude usage - the Agent SDK and `claude -p` headless mode - stops drawing from Pro/Max/Team/Enterprise subscription limits and instead draws from a separate monthly Agent SDK credit billed at standard API rates. Interactive Claude Code in the terminal is unaffected. For anyone running agents or automations on a Claude subscription, the prior 'programmatic loops at subscription pricing' arrangement ends - that usage is now metered at full API rates.
Predicted impact: 49-70% of paid users · ~$7-$350 per user
2026-04-15 · minor · nerf 18 · feature-gated
Anthropic moved newer Claude models to dateless model IDs (still pinned)
Anthropic stopped issuing dated snapshot IDs for newer models (the 4.6 generation onward) in favor of dateless IDs. Per Anthropic's model docs these dateless IDs are still pinned snapshots, not floating pointers - so version pinning is intact.
Predicted impact: 9-21% of paid users · ~$3-$30 per user
2025-09-17 · major · nerf 36 · model-swap
Anthropic admits to August 2025 Claude quality regression
After weeks of community complaints about Claude Code ignoring CLAUDE.md guidelines, Anthropic acknowledged a quality regression in August 2025.
Predicted impact: 35-70% of paid users · ~$4-$42 per user
2025-04-09 · info · nerf 0 · tier-added
Anthropic launches Claude Max plan at $100 and $200/mo - mirrors ChatGPT Pro
Anthropic introduced a two-level Claude Max plan on April 9, 2025: $100/mo for 5x Pro usage and $200/mo for 20x Pro usage. The tier later absorbed Claude Code access exclusively in April 2026.
Predicted impact: 1-3% of paid users · ~$0-$1 savings per user

Recommendations

Manageable risk. Continue normal usage with email alerts on new receipts.
Subscribe to receipts at gotnerfed.com/feed.xml or via Slack/Discord/Teams webhooks (paid tier).
Methodology: full scoring rubric at gotnerfed.com/methodology.