42Crunch for Few-shot poisoning test

A few-shot poisoning test targets data quality and model behavior when training examples are scarce. The goal is to measure whether small, carefully crafted changes to training data significantly alter model outputs, instructions, or extracted information. This assessment focuses on data-level manipulation, not on exploiting runtime endpoints.

middleBrick maps findings to OWASP API Top 10 (2023) and supports audit evidence for controls related to input validation and data exposure. The scanner detects issues such as prompt injection surfaces, encoding bypass attempts, and PII extraction that are relevant to few-shot poisoning test scenarios. It helps you prepare for compliance checks by surfacing findings that may align with security controls in SOC 2 Type II and PCI-DSS 4.0.

middleBrick performs LLM / AI Security scanning with 18 adversarial probes across three scan tiers. These probes include system prompt extraction, instruction override, DAN and roleplay jailbreaks, data exfiltration, cost exploitation, base64/ROT13 encoding bypass, translation-embedded injection, few-shot poisoning, markdown injection, multi-turn manipulation, indirect prompt injection, token smuggling, tool-abuse, nested instruction injection, and PII extraction. The scanner analyzes API definitions and runtime behavior to identify inputs that could be leveraged to bias or manipulate model learning, including suspicious parameter patterns and exposed sensitive fields.

middleBrick is a black-box scanner that requires only a URL. It works with any language, framework, or cloud setup without agents or SDKs. For authenticated scans, domain verification is enforced so that only the domain owner can submit credentials. The tool provides a web dashboard, CLI, GitHub Action, MCP Server, and API client for programmatic access. Scan results include prioritized findings and remediation guidance, with continuous monitoring options for ongoing risk tracking. Note that the scanner does not fix, patch, or block issues; it reports findings and suggests controls. It also does not perform active SQL injection or command injection testing, and it does not detect business logic vulnerabilities, which require domain-specific human review.

middleBrick does not detect blind SSRF or conduct intrusive exploit testing. It is not an auditor and cannot certify compliance with HIPAA, GDPR, ISO 27001, NIST, CCPA, or other regulations. The tool surfaces findings relevant to security and data integrity but should not replace a human pentester for high-stakes audits. Organizations should treat scanner output as one input to a broader evaluation of model behavior and data quality.