Confidential AI
Overview
Product details compiled from public sources, each with a citation.
Matrix Coverage
Where this product defends, by asset class and NIST CSF function. The Coverage column shows whether each asset is Primary, Secondary, or Adjacent to what the product does. The table omits empty rows and columns.
| Asset class | Protect | Coverage | Source |
|---|---|---|---|
| AI-Workload Platforms | Secondary | 1 | |
| AI Model | Secondary | 1 | |
| Training Data | Secondary | 1 | |
| Runtime AI Data | Primary | 1 | |
| AI Agent Identities | Secondary | 1 |
Framework Relevance
These frameworks include controls relevant to the asset classes Confidential AI defends. This is an editorial inference from the AI Defense Matrix asset-level crossmap, not a statement that Confidential AI implements these controls or is certified against them.
Expand Collapse
| Framework | Asset class | Relevant controls |
|---|---|---|
| NIST IR 8596 | AI-Workload Platforms | Containers, microservices, and libraries (AI-specific subset); inference endpoints (platform side) |
| AI Model | Models; Algorithms (model configuration) | |
| Training Data | Training data | |
| Runtime AI Data | Prompts (runtime); inference data | |
| AI Agent Identities | Agents as autonomous principals; Keys; Integrations and permissions | |
| CSA AI Controls Matrix | AI-Workload Platforms | Infrastructure Security; Threat & Vulnerability Management |
| AI Model | Model Security; Governance, Risk and Compliance | |
| Training Data | Data Security and Privacy Lifecycle Management; Model Security | |
| Runtime AI Data | Data Security and Privacy Lifecycle Management; Application and Interface Security | |
| AI Agent Identities | IAM; Governance, Risk and Compliance | |
| ISO 42001 | AI-Workload Platforms | A.6 AI system life cycle; A.4 Resources for AI systems |
| AI Model | A.6 AI system life cycle; A.10 Third-party and customer relationships; A.5 Assessing impacts of AI systems | |
| Training Data | A.7 Data for AI systems | |
| Runtime AI Data | A.7 Data for AI systems; A.8 Information for interested parties | |
| AI Agent Identities | A.9 Use of AI systems; A.3 Internal organization; A.5 Assessing impacts of AI systems | |
| Google SAIF | AI-Workload Platforms | Expand strong security foundations; secure and harden the AI deployment environment |
| AI Model | Protect the AI model; ensure model integrity, provenance, and weight security | |
| Training Data | Secure training data; data-security foundations; dataset provenance and integrity | |
| Runtime AI Data | Expand AI red-teaming; runtime input and output safety; prompt defense | |
| AI Agent Identities | Focus on Agents (explicit SAIF section); identity, authorization, and delegation controls | |
| SANS Critical AI Security Guidelines | AI-Workload Platforms | Conventional Security Controls (host AI within the existing ISMS; authentication and access controls; encryption at rest); AI Supply Chain Management (local vs. SaaS hosting trade-offs; internal model garden) |
| AI Model | Conventional Security Controls (protect model parameters with least privilege, encryption at rest, runtime obfuscation, and trusted execution environments); Data/Model Engineering Controls (adversarial training; alignment and fine-tuning); AI Supply Chain Management (public-model caution; transfer-attack exposure) | |
| Training Data | Conventional Security Controls (defend training data; avoid data commingling); Data/Model Engineering Controls (data-quality controls; poison-robust training); Data Minimization and Obfuscation (differential privacy; synthetic data; federated learning) | |
| Runtime AI Data | Model I/O Handling (sanitize, validate, and filter inputs and outputs; segregate user and system prompts; multilayered prompt-injection defense); Conventional Security Controls (protect augmentation and RAG data with vector-store access controls and validation); Data Minimization and Obfuscation (limit sensitive prompt content; context-window management); Limit Model Behavior (AI guardrails) | |
| AI Agent Identities | Secure Agentic Systems and AI Autonomy Controls (defined function scope; API and function-call gating; escalation and fallback); Limit Model Behavior (least-privilege focused functionality; human oversight; override capabilities) | |
| MITRE ATLAS | AI-Workload Platforms | AML.T0010 AI Supply Chain Compromise; AML.T0012 Valid Accounts (platform credential abuse); container and inference-server exploits |
| AI Model | AML.T0043 Craft Adversarial Data; AML.T0024 Exfiltration via AI Inference API (subtechniques: AML.T0024.001 Invert AI Model and AML.T0024.002 Extract AI Model); AML.T0018 Manipulate AI Model (integrity and backdoor) | |
| Training Data | AML.T0020 Poison Training Data; AML.T0019 Publish Poisoned Datasets; AML.T0024.000 Infer Training Data Membership | |
| Runtime AI Data | AML.T0051 LLM Prompt Injection; AML.T0054 LLM Jailbreak; AML.T0056 Extract LLM System Prompt | |
| AI Agent Identities | AML.T0053 AI Agent Tool Invocation; credential and delegation-chain abuse | |
| OWASP AI Exchange | AI-Workload Platforms | Development-time threats: supply chain attacks, model-platform CVEs, container escape |
| AI Model | Development-time and runtime model threats: model inversion, extraction, evasion, poisoning | |
| Training Data | Development-time threats: data poisoning, backdoor injection, dataset integrity violations | |
| Runtime AI Data | Input threats: prompt injection, adversarial inputs, evasion; runtime threats: RAG poisoning, memory tampering | |
| AI Agent Identities | Runtime threats: unauthorized agent actions, capability abuse, delegation chain exploitation | |
| OWASP LLM Top 10 | AI-Workload Platforms | LLM03 Supply Chain (compromised AI platform components); LLM04 Data and Model Poisoning (via platform) |
| AI Model | LLM03 Supply Chain; LLM04 Data and Model Poisoning; LLM09 Misinformation | |
| Training Data | LLM04 Data and Model Poisoning; LLM03 Supply Chain (dataset provenance) | |
| Runtime AI Data | LLM01 Prompt Injection; LLM02 Sensitive Information Disclosure; LLM08 Vector and Embedding Weaknesses; LLM05 Improper Output Handling | |
| AI Agent Identities | LLM06 Excessive Agency; LLM05 Improper Output Handling; unauthorized actions by AI agents | |
| OWASP Agentic Security Top 10 | AI-Workload Platforms | ASI04 Agentic Supply Chain Vulnerabilities (model and tool-platform components); ASI08 Cascading Failures (platform fault propagation) |
| AI Model | ASI04 Agentic Supply Chain Vulnerabilities (model provenance, weights, and dynamic loading) | |
| Training Data | ASI04 Agentic Supply Chain Vulnerabilities (dataset provenance and integrity) | |
| Runtime AI Data | ASI06 Memory & Context Poisoning; ASI01 Agent Goal Hijack (via prompt injection in runtime inputs) | |
| AI Agent Identities | ASI03 Identity and Privilege Abuse; ASI10 Rogue Agents; ASI09 Human-Agent Trust Exploitation; ASI02 Tool Misuse and Exploitation (when tied to agent permissions) |
Provenance
Last sourced 2026-07-16.
Expand Collapse
Sources
- Confidential AI
- “Customer prompts, responses, and model interactions are never visible.”
- “Protect proprietary model weights during inference, training, and fine-tuning.”
- “We run your AI workloads (inference, agents, training) in hardware-encrypted Trusted Execution Environments (TEEs).”
- “Tokens and API keys never exist in plaintext outside a TEE.”
- “Train on sensitive data and cryptographically prove exactly what data was used.”
- Confidential AI Pricing
Changelog
-
Added to the catalog from the Confidential AI documentation.
Found an error? Corrections are welcome. Suggest an edit.