# A vendor telemetry or incident-response source (Mandiant/GTIG, Microsoft, Unit 42, Huntress, watchTowr, Sysdig, CrowdStrike, Sophos) publishes, after issuance, an account of an attacker-operated AI agent framework used in a real intrusion that includes recovered agent artifacts (instruction, configuration or prompt files) and a described victim, for an actor other than JadePuffer/Storm-3168, on or before 2027-01-11.

- **Artifact ID:** SO-2026-006
- **Status:** OPEN
- **Public record:** https://record.cybersecurityhq.com/forecasts/so-2026-006
- **Machine-record SHA-256:** `a47e3c3b18ff8b20b3d342deee6574f019288c8d5004c5e0e5c844fd812bbe91`

## Complete structured record

```json
{
  "id": "SO-2026-006",
  "class": "EVIDENCE_PUBLICATION",
  "claim": "A vendor telemetry or incident-response source (Mandiant/GTIG, Microsoft, Unit 42, Huntress, watchTowr, Sysdig, CrowdStrike, Sophos) publishes, after issuance, an account of an attacker-operated AI agent framework used in a real intrusion that includes recovered agent artifacts (instruction, configuration or prompt files) and a described victim, for an actor other than JadePuffer/Storm-3168, on or before 2027-01-11.",
  "issued": "2026-10-13T12:30:00Z",
  "probability": 0.5,
  "benchmark": 0.4,
  "benchmark_kind": "ANALYST_PRIOR",
  "benchmark_basis": "Registry base: attacker-agent accounts Jul (Sysdig), Sep (GTIG, with artifacts), Sep (Microsoft, no artifacts): roughly one per six weeks, one in three with artifacts. The GTIG 8 Sep report pre-dates issuance and does not count.",
  "adjustment": 0.1,
  "resolution_rule": "YES if such a publication dated after issuance meets all four elements (agent framework, real intrusion, recovered artifacts, actor not JadePuffer).",
  "admissibility": "Only publications dated after 2026-10-13T12:30Z count; for EVENT-class, only events occurring after that timestamp count. Evidence pre-dating issuance resolves the forecast RETROSPECTIVE_DISCOVERY, not YES.",
  "exclusions": null,
  "resolves": "2027-01-11",
  "status": "OPEN",
  "outcome_date": null,
  "outcome_basis": null,
  "brier": null,
  "benchmark_brier": null,
  "resolution_source": "EXTERNAL",
  "experimental": true,
  "false_positive_cost": "Over-forecast would overstate how often vendors recover and publish agent artifacts.",
  "ledger_hash": "4390d1c0c82719f2",
  "revision_log": [],
  "record_links": []
}
```