「data-poisoning」の検索結果

Simulate supply chain and adversarial machine learning attacks by injecting poisoned data or targeted backdoors into training and fine-tuning datasets. Use this skill when assessing the integrity controls of MLOps pipelines or evaluating the resilience of AI models against highly targeted, stealthy manipulation intended to alter model behavior on specific triggers.

日本語の概要は準備中です。原文の説明を表示しています。

Execute and analyze AI Data Poisoning attacks. By subtly injecting malicious or targeted misinformation into an LLM's training or fine-tuning dataset, an attacker can covertly manipulate the model's future outputs, implant backdoors, or enforce biases without altering the model architecture.

日本語の概要は準備中です。原文の説明を表示しています。

Identify and simulate Data Poisoning attacks aimed at degrading or skewing an AI model's accuracy. This skill focuses on Adversarial Machine Learning concepts where attackers inject malicious or mislabelled data points into training or fine-tuning datasets (e.g., feedback loops) to bias the AI.

日本語の概要は準備中です。原文の説明を表示しています。

Identify poisoned training data and backdoored ML models across the pipeline using IBM's Adversarial Robustness Toolbox (activation clustering, spectral signatures, trigger reconstruction), Cleanlab for label-quality issues, and supply-chain checks like weight-hash verification and safetensors enforcement. Use before training or deploying on third-party/user-contributed data or downloaded checkpoints, during ML supply-chain reviews, or when investigating model misbehavior tied to specific inputs (suspected backdoor trigger).

日本語の概要は準備中です。原文の説明を表示しています。

Test RAG vector stores (Pinecone, Qdrant, Weaviate, Chroma, pgvector, FAISS) for embedding inversion, cross-tenant data leakage, and data poisoning per OWASP LLM08:2025. Use when performing an authorized security assessment of a RAG pipeline's retrieval layer or auditing multi-tenant vector-store isolation.

日本語の概要は準備中です。原文の説明を表示しています。

This skill should be used when the user asks to "scan AI systems for security threats", "check for prompt injection vulnerabilities", "assess model security posture", "detect data poisoning risks", or "audit AI/ML pipeline security".

日本語の概要は準備中です。原文の説明を表示しています。

Exploit AI applications using Indirect Prompt Injection. This skill focuses on hiding malicious instructions within data sources (web pages, documents, emails) that the LLM processes, causing the AI to execute unintended actions or leak data without direct user interaction.

日本語の概要は準備中です。原文の説明を表示しています。

Test for indirect prompt injection vulnerabilities where malicious instructions are injected through external data sources (websites, emails, documents, database records) that the LLM processes. Use this skill when assessing LLM-integrated applications that process user-generated content, retrieve web pages, parse emails, or read documents. Covers injection via web content, email bodies, user profiles, and database records that are fed to LLM context.

日本語の概要は準備中です。原文の説明を表示しています。

Identify poisoned training data and backdoored models across the ML pipeline.

日本語の概要は準備中です。原文の説明を表示しています。

Test vector stores for embedding inversion, cross-tenant leakage, and poisoning.

日本語の概要は準備中です。原文の説明を表示しています。