anthropic/claude-haiku-4.5 · Model identity and integrity report

Family

anthropic

V3 sub-model reference

Available ✓

Checks

知識截止 · 能力探針 · 拒絕模板 · 拒絕溫度參數

Family-fingerprint coverage

60 weighted rules · 8 feature categories

Concrete reference answers, refusal templates, and individual signal weights are not published because disclosure would help spoofing endpoints evade detection.

98 runs · 56 hosts · 31 anomalies

Family-fingerprint consistency

  anthropic       87  ██████████████████████░░░   89% (expected)
  unknown          7  ██░░░░░░░░░░░░░░░░░░░░░░░    7% (suspicious)
  openai           3  █░░░░░░░░░░░░░░░░░░░░░░░░    3% (suspicious)
  google           1  ░░░░░░░░░░░░░░░░░░░░░░░░░    1% (suspicious)

Observed V3 sub-model mapping

  → anthropic/claude-haiku-4.5              53 (54%) ✓ correct sim=0.95
  → abstain                                 26 (27%) — low confidence
  → unknown                                  8 ( 8%) ⚠ mismatch
  → anthropic/claude-opus-4.5                6 ( 6%) ⚠ mismatch sim=0.68
  → anthropic/claude-sonnet-4.5              1 ( 1%) ⚠ mismatch sim=0.83
  → anthropic/claude-fable-5                 1 ( 1%) ⚠ mismatch sim=0.85
  → anthropic/claude-sonnet-4.6              1 ( 1%) ⚠ mismatch sim=0.66
  → anthropic/claude-opus-4.7                1 ( 1%) ⚠ mismatch sim=0.76
  → openai/gpt-5.4                           1 ( 1%) ⚠ mismatch sim=1.00

Capability score distribution (auxiliary)

  0-49     ████████████░░░░░░░░░░░░░   46
  50-69    █░░░░░░░░░░░░░░░░░░░░░░░░    3
  70-89    ███████████░░░░░░░░░░░░░░   43
  90-100   ██░░░░░░░░░░░░░░░░░░░░░░░    6

BazaarLink Probe combines a Bayesian V2 family classifier across nine fingerprint categories with V3 sub-model references to verify endpoints claiming anthropic/claude-haiku-4.5. Low family consistency, a different V3 match, or divergence from selfClaim is flagged as anomalous. BazaarLink Probe home.