Shareable report · ACRIP-XCORR-20261102-B0C70C26

PRODUCTIONNeon main٠٧:٠٢ م

النتائج المتحققة

لا يوجد

UNKNOWN: Anthropic documents stolen AI API keys/session tokens as targets, loot and attack compute; GTG-50020 caused an AI vendor sandbox to disclose production keys, switched workloads to stolen keys and then targeted roughly 30 AI companies in about four days.
حد الدليل: GTG-50020 identity behind Black October or universal AI-driven efficacy uplift.
UNKNOWN: Cross-source revalidation confirms a concentrated October wave against AI developer platforms with repeated environment-variable/API-credential targeting; tracked downstream Bedrock abuse reached ~2.5M calls and >$100k list-price usage in two cases.
حد الدليل: One actor, one exploit, one vulnerability or identical initial-access chain across all vendors.
UNKNOWN: Wiz observed 23.106.46.239, 38.246.114.118, 104.243.32.50 and 31.57.201.201 testing stolen keys; linked the first to momo/LLM resale and the second to Xingjiabi. Zeabur independently lists 23.106.46.239 in incident activity and 31.57.201.201 in earlier root sign-ins while leaving the September→October relation unconfirmed.

حد الدليل: That the IPs performed every initial compromise, or that same infrastructure implies same human/operator.

OPEN [HIGH]: PLATFORM_ROOT_CAUSE_AND_MARKET_DENOMINATOR — No authoritative cross-vendor root-cause map or denominator exists for compromised AI credentials, brokers/resellers, downstream buyers, realized spend or actor overlap.
OPEN [HIGH]: STOLEN_COMPUTE_OPERATOR_LINEAGE — Initial-access actors, validators, brokers, resellers and final AI-compute consumers are not resolved as same vs separate entities.
OPEN [MEDIUM-HIGH]: SHARED_UPSTREAM_ACCESS_LINEAGE — No public evidence establishes a shared upstream credential compromise, publisher account, build system, dependency or release authority linking distinct SCSE incidents to AI-platform intrusions.
OPEN [HIGH]: INFRASTRUCTURE_CONTROL_AND_ACTOR_CROSSWALK — Ownership/control and operator identity for momo/Xingjiabi/validator nodes, proxy pools and original platform intrusions are not resolved for single-campaign attribution.