MetaDefender® Platform Guardrails for LLMs
Secure files, content, and data before they reach your LLM. Stop file level threats, reduce RAG poisoning risk, enforce one-way transfer, and protect your AI investment.
- Prevention First Security
- ディープファイルサニタイゼーション
- Hardware Enforced Isolation
OPSWAT 信頼
The New AI Attack Surface
File-Borne Malware in AI Pipelines
Malicious payloads hide inside common business files like PDFs, Office documents, and archives. Typical AI guardrails focused on text do not neutralize embedded file-level risks.
Knowledge Manipulation Through Untrusted Documents
Malicious or deceptive documents can enter retrieval pipelines, get indexed, and silently influence model outputs over time, turning the knowledge base into an attack vector.
Sensitive Data Exposure to AI Systems
Users upload financial data, source code, credentials, and customer records into AI workflows. Once exposed to external models or poorly governed services, organizations face loss of data control and potential regulatory liability.
Embedded Prompt Injection Inside Files
Instructions hidden inside uploaded documents, rather than typed directly into chat, can manipulate model behavior and downstream tools when retrieved through RAG or agent workflows.
Unknown and Zero-Day File-Based Threats
AI workflows encourage massive content ingestion, increasing exposure to previously unseen threats. Detection alone is not enough. Prevention must occur before content enters the pipeline, or organizations risk regulatory penalties and reputational harm from undetected breaches.
Prevention First Security for Enterprise AI
MetaDefender Core applies a prevention-first model to AI content flows and secures what enters the model, what gets indexed, and what crosses trust boundaries.
File Sanitization and Threat Removal
Strips embedded objects and out-of-policy content, and regenerates safe, usable files. Neutralizes both known and unknown malware without relying on signature-based detection.
Secure RAG and Knowledge Pipelines
Ensures only trusted, policy-approved content is indexed into retrieval systems and vector stores, reducing RAG poisoning risk and long-lived knowledge manipulation.
Sensitive Data Control
Enforces what content is allowed into public LLMs, internal copilots, and external AI APIs, scanning for PII, PHI, credentials, and financial data using OCR-powered hidden text detection.
Policy-Driven Enforcement and Quarantine
Organizations define what content is permitted, what must be sanitized, and what is blocked or quarantined, creating a true control layer for enterprise AI content handling.
Hardware-Enforced One-Way Transfer (Optical Diode)
For high-assurance environments, MetaDefender Optical Diode™ provides a hardware-enforced, unidirectional data path with no return channel, preventing data exfiltration even if downstream systems are compromised.
特徴
エンジンの数が多いほど、検出力は高まります。
- ほぼ100%のマルウェアを検出
- 30種以上の主要アンチウイルスエンジンで同時スキャンを実行
他では見逃す脅威も確実に除去
- 200種類以上のファイル形式に対応
- 多層構造の圧縮ファイルも階層的に無害化
- 安全性を確保したうえでファイルを再生成
セキュリティ・クリティカルなワークフローのための真のファイルタイプ検出
- AI強化
- 数ミリ秒で偽装されたファイル形式を検出します
- パフォーマンスを損なうことなくインラインで適用する
機密データの漏えいを未然に防止します。
- AIモデルにより、非構造化テキストを事前定義されたカテゴリへ自動分類
- 125種類以上のファイル形式において、PII、PHI、PCIなどの特定された機密情報を自動的にマスキングします
- 画像内の文字を認識するOCR（光学式文字認識）機能をサポート
ゼロデイ脅威に対して、圧倒的な防御力を発揮
- 高速でファイルを分析する
- 回避防止サンドボックスエンジンがIOCを抽出
- ゼロデイ脅威の特定
- API またはローカル統合により、マルウェアの詳細な分類が可能
リアルタイムのThreat Intelligence検出の強化
- 50B以上のアーティファクトにわたって、グローバルなIOC、IP、URL、ファイルのレピュテーションを関連付けます。
- 新たな脅威を迅速に阻止
- 下流分析の充実
ソフトウェアサプライチェーンのリスクを可視化・保護します。
- OSS（オープンソースソフトウェア）やサードパーティ製コンポーネント、依存関係に伴うリスクを管理
- コードベースの透明性、セキュリティ、コンプライアンスを確保
インストール前に脆弱性を検知し、リスクを未然に防止
- インストール前にソフトウェアの既知の脆弱性を検査
- デバイス非稼働時にも脆弱性スキャンを実行
- 稼働中のアプリケーションおよび関連ライブラリを迅速に分析
Enable Instant Detection of a File’s Geographic Source
- PE、MSI、SFX（自己解凍型アーカイブ）などのアップロードファイルから地理的出所を検出
- ハッシュ情報やメタデータを自動解析し、制限対象の国やベンダーを特定
深くネストされたアーカイブファイルを再帰的に抽出して分析する
- 設定可能な深さまで再帰的に抽出する
- すべてのエンジンにわたるシングルパス抽出
- アーカイブ爆弾の検知と封じ込め
- 暗号化およびパスワード保護されたアーカイブのサポート
![OPSWAT 技術イメージ]()
ファイルタイプの検出セキュリティ・クリティカルなワークフローのための真のファイルタイプ検出
- AI強化
- 数ミリ秒で偽装されたファイル形式を検出します
- パフォーマンスを損なうことなくインラインで適用する
99%以上の精度偽装エクステンションについて![OPSWAT 技術イメージ]()
脅威の検出リアルタイムのThreat Intelligence検出の強化
- 50B以上のアーティファクトにわたって、グローバルなIOC、IP、URL、ファイルのレピュテーションを関連付けます。
- 新たな脅威を迅速に阻止
- 下流分析の充実
高速トリアージにかかる総時間を短縮する透明な重要な環境をより明確に保護する![OPSWAT 技術イメージ]()
アーカイブ抽出深くネストされたアーカイブファイルを再帰的に抽出して分析する
- 設定可能な深さまで再帰的に抽出する
- すべてのエンジンにわたるシングルパス抽出
- アーカイブ爆弾の検知と封じ込め
- 暗号化およびパスワード保護されたアーカイブのサポート
160種類以上のアーカイブ形式サポート
導入オプション
Cloud Native
Deploy MetaDefender Core in your cloud environment for scalable, on-demand AI pipeline protection.
Integrates with cloud-based AI workflows via REST API, supporting elastic scaling for variable file ingestion volumes across LLM applications and RAG pipelines.
オンプレミス
Full on-premises deployment for organizations requiring complete control over data and infrastructure.
Air-Gapped / High-Assurance
Air-gapped deployment with MetaDefender Optical Diode for hardware-enforced unidirectional data transfer.
統合
It scans at every stage, from file upload portals and RAG ingestion pipelines to CI/CD workflows used in AI model and chatbot development. The platform connects to existing enterprise AI environments, including cloud platforms such as AWS and Azure, without requiring changes to application logic or model infrastructure.
Where MetaDefender Core
Fits in the AI Stack
MetaDefender Core acts as the AI security gateway, inspecting and sanitizing content before file upload, before RAG ingestion, before tool execution, and before data crosses a trust boundary.
Protect AI Copilots Handling Sensitive Financial Data
Financial institutions using LLM-powered copilots for research, compliance, and customer service need to prevent sensitive data leakage and ensure that uploaded documents are free of embedded threats. Proactive DLP and Deep CDR™ Technology enforce content-level controls before files reach the model.
High-Assurance AI with Hardware-Enforced Isolation
Government and defense agencies require the highest levels of data assurance. MetaDefender Core sanitizes all content entering classified or sensitive AI environments, and MetaDefender Optical Diode ensures no data can flow back through the ingestion path — meeting strict cross-domain transfer requirements.
Secure AI-driven Analytics in Operational Environments
Manufacturers using AI for predictive maintenance, quality control, and supply chain optimization must protect against file-borne threats entering through data ingestion. MetaDefender Core provides policy-driven enforcement at every ingestion point, with air-gapped deployment options for isolated OT networks.
Secure AI Deployments Across OT and IT Environments
Energy and utilities organizations deploying AI for operational intelligence need to ensure that untrusted files and data feeds cannot introduce malware or manipulate models connected to operational technology networks. MetaDefender Optical Diode enforces one-way data transfer between IT and OT zones.
- 金融
金融Protect AI Copilots Handling Sensitive Financial Data
Financial institutions using LLM-powered copilots for research, compliance, and customer service need to prevent sensitive data leakage and ensure that uploaded documents are free of embedded threats. Proactive DLP and Deep CDR™ Technology enforce content-level controls before files reach the model.
- 政府政府
High-Assurance AI with Hardware-Enforced Isolation
Government and defense agencies require the highest levels of data assurance. MetaDefender Core sanitizes all content entering classified or sensitive AI environments, and MetaDefender Optical Diode ensures no data can flow back through the ingestion path — meeting strict cross-domain transfer requirements.
- 製造業製造業
Secure AI-driven Analytics in Operational Environments
Manufacturers using AI for predictive maintenance, quality control, and supply chain optimization must protect against file-borne threats entering through data ingestion. MetaDefender Core provides policy-driven enforcement at every ingestion point, with air-gapped deployment options for isolated OT networks.
- エネルギー＆ユーティリティエネルギー＆ユーティリティ
Secure AI Deployments Across OT and IT Environments
Energy and utilities organizations deploying AI for operational intelligence need to ensure that untrusted files and data feeds cannot introduce malware or manipulate models connected to operational technology networks. MetaDefender Optical Diode enforces one-way data transfer between IT and OT zones.
おすすめ資料
AIデータパイプラインとLLMを活用したアプリケーションのセキュリティ確保
AI加速型Software ファイルセキュリティをサイバー防衛の新たな最前線とする
よくある質問
MetaDefender Core supports over 200 file types including PDFs, Office documents, archives, images, media files, source code, and executables, covering the full range of content commonly ingested by enterprise AI systems.
Deep CDR™ Technology does not rely on detecting known threats. It strips all active content from files and reconstructs clean, usable versions, neutralizing both known and unknown malware, including zero-day threats.
Yes. MetaDefender Core inspects and sanitizes files before they are indexed into vector stores or retrieval systems, reducing the risk of RAG poisoning and long-term knowledge manipulation.
The MetaDefender Optical Diode is a hardware-enforced, one-way data transfer device. It physically prevents data from flowing back into a protected environment — required for defense, critical infrastructure, and any deployment where software-only controls are insufficient.
MetaDefender Core integrates via REST API or ICAP at any data ingestion point, including file upload portals, RAG pipelines, CI/CD workflows, and AI training data feeds. No changes to application logic or model infrastructure are required.
Yes. MetaDefender Core provides secure input validation, complete audit trails, file hashing, and logging that support compliance with the EU AI Act, Cyber Resilience Act, GDPR, HIPAA, and other emerging AI regulatory frameworks.
Yes. Proactive DLP scans for PII, PHI, financial data, and credentials. It also uses OCR to detect and redact hidden text within images and visual content that could bypass human review.
MetaDefender Core deploys cloud-native, on-premises, or in air-gapped architectures. For high-assurance environments, it pairs with the Optical Diode for hardware-enforced unidirectional transfer.
