v0.5.0 — Prompt Injection Scanning

What's New

Prompt Injection Scanning

Optional AI-powered scanning of uploaded documents for prompt injection attacks. When enabled, files that pass GuardDuty malware scanning are additionally scanned using the protectai/deberta-v3-base-prompt-injection-v2 ONNX model before reaching the egress bucket.

Supported formats: PDF, DOCX, PPTX, and plain text (TXT, CSV, MD, JSON, XML, HTML, YAML, LOG). Non-text files pass through with score 0.

How it works: The file router invokes a container-image Lambda synchronously. Text is extracted, tokenized into overlapping 512-token chunks, and scored by the model. Files scoring above the configurable threshold (default 80) are quarantined with an SNS alert.

Usage

module "secure_upload" {
  source = "gofastercloud/secure-upload/aws"

  name_prefix                      = "myapp"
  enable_prompt_injection_scanning = true

  # Optional tuning
  prompt_injection_threshold = 80   # 0-100, default 80
  prompt_injection_memory_size = 3008  # more memory = more CPU = faster cold starts
}

New Variables

Variable	Default	Description
enable_prompt_injection_scanning	false	Enable the scanning pipeline
prompt_injection_threshold	80	Score (0-100) above which files are quarantined
prompt_injection_memory_size	3008	Scanner Lambda memory in MB
prompt_injection_timeout	120	Scanner Lambda timeout in seconds
prompt_injection_reserved_concurrency	5	Scanner Lambda concurrency
prompt_injection_image_uri	null	BYO pre-built container image

Other Changes

• File router timeout auto-adjusts to scanner_timeout + 30s when scanning is enabled
• ECR repository with KMS encryption and lifecycle policy for scanner images
• 7 new validation tests (49 total)
• Documentation updated (README, ARCHITECTURE.md, CHANGELOG.md)

Full Changelog: v0.4.0...v0.5.0