AIセキュリティマップにマッピングされた外部作用的側面における負の影響「AIにより作成したメールや音声により機密情報の入力を促し、窃取」のセキュリティ対象、それをもたらす攻撃・要因、および防御手法・対策を示しています。
セキュリティ対象
- 非消費者
攻撃・要因
- ソーシャルエンジニアリング攻撃
- 可用性の悪用
- 精度の悪用
防御手法・対策
AIシステムの開発フェーズにおける防御手法
1. データ収集・前処理
- 匿名化技術
- 差分プライバシー
- 暗号化技術
2. モデルの選定・学習・検証
- 差分プライバシー
- 連合学習
- マシン・アンラーニング
- 暗号化技術
- 生成AI向け電子透かし
3. システムの実装
- 生成AI向け電子透かし
4. システムの提供・運用・保守
5. システムの利用
- AIによる出力の識別
- 偽情報の検出
- ディープフェイクの検知
参考文献
ソーシャルエンジニアリング攻撃
匿名化技術
差分プライバシー
- Deep Learning with Differential Privacy, 2016.0
- Semi-supervised Knowledge Transfer for Deep Learning from Private Training Data, 2017.0
- Learning Differentially Private Recurrent Language Models, 2018.0
- Efficient Deep Learning on Multi-Source Private Data, 2018.0
- Evaluating Differentially Private Machine Learning in Practice, 2019.0
- Tempered Sigmoid Activations for Deep Learning with Differential Privacy, 2020.0
連合学習
- Practical Secure Aggregation for Federated Learning on User-Held Data, 2016.0
- Communication-Efficient Learning of Deep Networks from Decentralized Data, 2017.0
- Federated Learning: Strategies for Improving Communication Efficiency, 2018.0
- Federated Optimization in Heterogeneous Networks, 2020.0
- SCAFFOLD: Stochastic Controlled Averaging for Federated Learning, 2020.0
- Federated Learning with Matched Averaging, 2020.0
マシン・アンラーニング
- Making AI Forget You: Data Deletion in Machine Learning, 2019.0
- Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep Networks, 2020.0
- Certified Data Removal from Machine Learning Models, 2020.0
- Forgetting Outside the Box: Scrubbing Deep Networks of Information Accessible from Input-Output Observations, 2020.0
- Approximate Data Deletion from Machine Learning Models, 2021.0
- Fast Yet Effective Machine Unlearning, 2021.0
- Machine Unlearning for Random Forests, 2021.0
- Machine Unlearning of Features and Labels, 2023.0
暗号化技術
- Gazelle: A Low Latency Framework for Secure Neural Network Inference, 2018.0
- Faster CryptoNets: Leveraging Sparsity for Real-World Encrypted Inference, 2018.0
- nGraph-HE2: A High-Throughput Framework for Neural Network Inference on Encrypted Data, 2019.0
- Privacy-Preserving Machine Learning with Fully Homomorphic Encryption for Deep Neural Network, 2021.0
生成AI向け電子透かし
AIによる出力の識別
- Defending Against Neural Fake News, 2019.0
- Real or Fake? Learning to Discriminate Machine from Human Generated Text, 2019.0
- Automatic Detection of Generated Text is Easiest when Humans are Fooled, 2020.0
- DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature, 2023.0
- Inspection and Control of Self-Generated-Text Recognition Ability in Llama3-8b-Instruct, 2025.0
偽情報の検出
- Fake News Detection on Social Media: A Data Mining Perspective, 2017.0
- CSI: A Hybrid Deep Model for Fake News Detection, 2017.0
- Towards Few-Shot Fact-Checking via Perplexity, 2021.0
- Fact-Checking Complex Claims with Program-Guided Reasoning, 2023.0
- Towards LLM-based Fact Verification on News Claims with a Hierarchical Step-by-Step Prompting Method, 2023.0
ディープフェイクの検知
- Two-Stream Neural Networks for Tampered Face Detection, 2017.0
- Exposing DeepFake Videos By Detecting Face Warping Artifacts, 2019.0
- Exposing Deep Fakes Using Inconsistent Head Poses, 2019.0
- CNN-generated images are surprisingly easy to spot… for now, 2020.0
- Face X-ray for More General Face Forgery Detection, 2020.0
- FakeCatcher: Detection of Synthetic Portrait Videos using Biological Signals, 2020.0
- End-to-end anti-spoofing with RawNet2, 2021.0