T1700 · exfiltration · ATT&CK 20.0

AI-Enabled Exfiltration

Adversaries may utilize [Direct Prompt Injection](https://attack.mitre.org/techniques/T1699), [Indirect Prompt Injection](https://attack.mitre.org/techniques/T1698) or other methods to instruct enterprise AI agents to exfiltrate data.(Citation: Prompt-Injection Exfil)(Citation: Anthropic Report Sep 2026) When enterprise systems incorporate the use of AI assistants or other AI agents, adversaries may abuse these systems to weaponize sanctioned tools available to the agent.(Citation: Google Security State of Prompt Injections) As an AI agent processes queries or ingests data, an injected prompt can cause the agent to use its tool connections in unauthorized or unintended ways. When AI agents are induced to exfiltrate data, other exfiltration techniques likely apply as well to transfer the information out of the network, such as [Exfiltration to Code Repository](https://attack.mitre.org/techniques/T1567/001) or [Exfiltration Over Web Service](https://attack.mitre.org/techniques/T1567).(Citation: Google Cloud AI Risk Sep 2026)(Citation: Barracuda Prompt Injection Phishing Emails)

Open in the interactive knowledge mesh

Detection overview

Use the published detection strategies below and validate required telemetry in the target environment.

Observed groups

No current Enterprise group relationship is published in this bundle.

Cyber Knowledge context

Use these routes to move from the ATT&CK behavior into explanation, implementation, evidence handling, validation, and defensive operations. Relevance is generated from explicit identifiers/names and governed topic mappings; it is not attribution evidence.

AI red teaming, evaluation, and reproducible security testing
AI Security · Governed topic match · 54/100Windows endpoint and identity forensics
Digital Forensics & Incident Response (DFIR) · Tactic learning route · 24/100Module 6 — Alert triage, investigation, and escalation
Blue Team & Defensive Security · Tactic learning route · 24/100Infrastructure as code, CI/CD, artifact provenance, and policy as code
Cloud Security · Tactic learning route · 24/100

MITRE mitigations

M1017 · User Training

User Training involves educating employees and contractors on recognizing, reporting, and preventing cyber threats that rely on human interaction, such as phishing, social engineering, and other manipulative techniques. Comprehensive training programs create a human firewall by empowering users to be an active component of the organization's cybersecurity defenses. This mitigation can be implemented through the following measures: Create Comprehensive Training Programs: - Design training modules tailored to the organization's risk profile, covering topics such as phishing, password management, and incident reporting. - Provide role-specific training for high-risk employees, such as helpdesk staff or executives. Use Simulated Exercises: - Conduct phishing simulations to measure user susceptibility and provide targeted follow-up training. - Run social engineering drills to evaluate employ…

MITRE mitigation source
M1048 · Application Isolation and Sandboxing

Application Isolation and Sandboxing refers to the technique of restricting the execution of code to a controlled and isolated environment (e.g., a virtual environment, container, or sandbox). This method prevents potentially malicious code from affecting the rest of the system or network by limiting access to sensitive resources and critical operations. The goal is to contain threats and minimize their impact. This mitigation can be implemented through the following measures: Browser Sandboxing: - Use Case: Implement browser sandboxing to isolate untrusted web content and prevent malicious web pages or scripts from accessing sensitive system resources or initiating unauthorized downloads. - Implementation: Use browsers with built-in sandboxing features (e.g., Google Chrome, Microsoft Edge) or deploy enhanced browser security frameworks that limit the execution scope of active content.…

MITRE mitigation source
M1022 · Restrict File and Directory Permissions

Restricting file and directory permissions involves setting access controls at the file system level to limit which users, groups, or processes can read, write, or execute files. By configuring permissions appropriately, organizations can reduce the attack surface for adversaries seeking to access sensitive data, plant malicious code, or tamper with system files. Enforce Least Privilege Permissions: - Remove unnecessary write permissions on sensitive files and directories. - Use file ownership and groups to control access for specific roles. Example (Windows): Right-click the shared folder → Properties → Security tab → Adjust permissions for NTFS ACLs. Harden File Shares: - Disable anonymous access to shared folders. - Enforce NTFS permissions for shared folders on Windows. Example: Set permissions to restrict write access to critical files, such as system executables (e.g., `/bin` or `…

MITRE mitigation source

MITRE detection strategies and analytics

DET0927 · Detect AI-Enabled Exfiltration
  • AN2079 · Analytic 2079 — An analyst observes an AI-associated process access data outside $AuthorizedAIDataScope$, followed within $EndpointExfilWindow$ by transfer-capable execution and an outbound connection to a destination outside $AuthorizedAIEndpoints$ from the same process lineage.
  • AN2078 · Analytic 2078 — An analyst observes an AI agent access $ProtectedDataScope$, invoke a tool/API using that information or a derived payload, and the destination service record a successful transfer to a recipient, repository, tenant, domain, or service outside $AuthorizedAIDataFlows$.
  • AN2080 · Analytic 2080 — An analyst observes an AI-associated process access data outside $AuthorizedAIDataScope$, followed within $EndpointExfilWindow$ by transfer-capable execution and an outbound connection from the same lineage to a destination outside $AuthorizedAIEndpoints$.
MITRE detection source