Net Good IndexSubmit a correction

Harm Ledger · provisional · Economics & Productivity

Anthropic disrupts Claude-assisted ShinyHunters theft and extortion campaigns

Anthropic’s September 2026 threat report says financially motivated operators suspected as ShinyHunters affiliates used Claude Haiku, Sonnet, and Opus models to steal data at scale—including more than a terabyte from one technology provider, millions of payment-card records, and tens of millions of airline passenger records—then extort victims.

10 Sep 2026Tier 3 Major HarmMethodology 0.1

Current score

−0.88

10 base · Major Harm (tier 3 of 5, 10 pts)
× 0.5000 attribution · Material acceleration
× 0.5000 evidence · External expert evaluation
× 0.7000 realization · Independently validated or deployed
× 0.5000 durability
Event-level product before credit split: 0.88

Large-scale theft and extortion of corporate and consumer data is major harm (tier 3). Human operators ran the campaigns with material Claude uplift (0.50). Evidence is a first-party threat-intel report, not court findings (0.50). Realization is reported completed theft, later disrupted (0.70).

What happened

The report covers activity disrupted between December 2025 and August 2026. Anthropic describes AI-orchestrated credential harvesting, database theft, a payment-card storefront, and pay-or-leak pressure, including a public staging of stolen data. Humans set targets and ran the crime; Claude was used across reconnaissance, exploitation, and data processing. Anthropic says it disrupted the observed clusters. Independent victim confirmation of the largest claimed volumes has not been published in this ledger’s sources.

Model attribution

Claude
Version unspecified
−0.88

Claude

Haiku, Sonnet, and Opus models Anthropic says financially motivated affiliates used to harvest credentials, steal databases, and support extortion.

Anthropic names Claude Haiku/Sonnet/Opus and excludes Fable and Mythos; no single SKU is identified, so credit stays at version-unspecified Claude.

Attribution 0.5000 · Credit share 100% · Anthropic

Claims

  • Anthropic reports disrupting Claude-assisted financially motivated campaigns that stole large corporate datasets and used them for extortion.

    outcome · supported

  • Claude Fable or Mythos carried out these thefts autonomously without human operators.

    attribution · disputed

Sources

primary sources

Secondary domains: Safety, Computer Science

Revision history

  • 13 Sep 2026 · 0.00 → 0.88

    Imported events under methodology 0.1.