Vega launches Vega II — purpose-built agentic cyber defense platform with first security-trained model, agentic memory, and data gateway

On Sept 30–Oct 1, 2026, Vega (post-stealth agentic-SOC startup, $185M raised) introduced Vega II, its largest release since stealth: a proprietary open-weight AI model trained specifically for cyber defense that runs alongside frontier models to handle detection/triage/investigation verdicts, Vega Memory (persistent environment ground-truth that outlasts tickets), and Vega Gateway (ingests any source into object storage via OpenTelemetry/webhook/API without legacy SIEM pipelines).

It is the first vendor to ship a security-domain-specialized model plus persistent agentic memory for SOC operations, directly attacking the two pain points CISOs cite with agentic SOCs: evidence unreachability and context loss across cases. Signals that agentic SOC differentiation is moving beyond LLM wrappers to purpose-trained models — a meaningful architectural shift in the AI-for-Security market.

Mid-to-large SOC teams evaluating agentic/Copilot-style security operations should assess Vega II against generic-LLM SOC assistants; watch benchmark results on the specialized model's verdict accuracy and the memory layer's auditability before adoption.