New attack technique: Invoke Vertex AI Model (gcp.impact.invoke-vertex-ai-model)

[Minosity-VR]

What does this PR do?

Adds a new GCP attack technique gcp.impact.invoke-vertex-ai-model that simulates an attacker invoking a Gemini generative AI model via the Vertex AI API to run LLM workloads and incur unexpected costs.

Motivation

GCP parity effort with existing AWS techniques (analogous to aws.impact.bedrock-invoke-model).

Checklist

• The attack technique emulates a single attack step, not a full attack chain
• We have factual evidence & references that the attack technique was used by real malware, pentesters, or attackers
• The attack technique makes no assumption about the state of the environment prior to warming it up

Test results

• Technique detonates successfully
• Detection log identified (google.cloud.aiplatform.v1.PredictionService.GenerateContent) — there is no audit log emitted for Vertex AI model invocations without non-default Data Access audit log configuration; a single invocation detection would yield too many false positives