Low-confidence report · Case No. FB-2026-09-30-anthropic-inc-gentlemen-ransomware-opencode
Ransomware affiliates of INC and The Gentlemen ran their intrusions through OpenCode on Claude Opus 4.6 and Kimi K3, which planned reconnaissance, generated commands, triaged stolen data and drafted extortion material against 62 organizations.
This report is hidden because you're viewing high- and medium-confidence reports only. It isn't counted in the scores you're seeing.
Leaderboard / Case No. FB-2026-09-30-anthropic-inc-gentlemen-ransomware-opencode
Ransomware affiliates of INC and The Gentlemen ran their intrusions through OpenCode on Claude Opus 4.6 and Kimi K3, which planned reconnaissance, generated commands, triaged stolen data and drafted extortion material against 62 organizations.
Score breakdown
| Component | Value | Reasoning |
|---|---|---|
| Sentence-Years | 11 | 18 USC 1030(a)(2)(C), 18 USC 1030(a)(7)(C), 18 USC 1030(a)(5)(A). Stealing victims' financial and business data for extortion is 18 USC 1030(a)(2)(C) at the five-year level for private financial gain; the extortion demands (e.g. a EUR 15M demand to Aidon) are 1030(a)(7)(C) (five years); the encryptor running on a victim workstation is 1030(a)(5)(A), kept at the one-year misdemeanor maximum because no loss figure is reported. |
| × Contribution | 2 | Ran the operation: SOCRadar describes an "AI-executed, human-approved operating structure" in which the models planned reconnaissance, selected techniques, generated commands, debugged tooling, triaged stolen data and drafted extortion material, waiting for the operator's approval before each step. It does not say whether the models executed commands directly, so operated is the highest level the source supports, not a certainty. |
| × Blast Radius | 2 | Government / critical infrastructure: SOCRadar says three INC leak-site victims, including the City of Princeton, Texas (a US municipal government), match the operators' internal case files; the other named victims are private companies. |
| × Legal status | 1 | Crime. |
| + Tradecraft | 3 | Coverage of SOCRadar's report says the AI was used to analyse stolen credentials. Lateral movement, persistence and EDR-killing tools appear in the operation, but the sources don't attribute them to the models, so they are left out. |
| + Guardrails | 5 | Jailbroken: SOCRadar says that when a model refused, the operator rewrote the request under an "Authorized Red Team Assessment Request" template rather than switching models, so the models were talked past their safeguards. The "WormGPT" label was only a persona prompt, not an uncensored model. |
| + Dwell Time | 3 | SOCRadar gives no dates for the AI sessions, only an operation running from mid-July to live encryption on 30 September 2026, so date_precision is before with 30 September as the latest possible date. The exposure came to light in late September, so dwell from that latest date is 0 days, detected by a third party (SOCRadar). |
| Total | 55 |
Charges the human could face
18 USC 1030(a)(2)(C)Up to 5 years18 USC 1030(a)(7)(C)Up to 5 years18 USC 1030(a)(5)(A)Up to 1 year