Lateral Movement
Lateral movement is when an intruder who has already broken into one part of your computer system tries to sneak into other, more sensitive areas of your AI setup to gain more control.
Lateral movement refers to the techniques an adversary uses to pivot from an initial point of compromise to other components within the AI ecosystem, such as model registries, training pipelines, or vector databases, to escalate privileges or access sensitive data.
The adversary is trying to move through your AI environment. Lateral Movement consists of techniques that adversaries may use to gain access to and control other systems or components in the environment. Adversaries may pivot towards AI Ops infrastructure such as model registries, experiment trackers, vector databases, notebooks, or training pipelines. As the adversary moves through the environment, they may discover means of accessing additional AI-related tools, services, or applications. AI agents may also be a valuable target as they commonly have more permissions than standard user accounts on the system.
evolution
- 2018 · historyMITRE ATT&CK Framework Expansion
MITRE formally codified 'Lateral Movement' (TA0008) as a core tactic in the ATT&CK framework, establishing the standard terminology for post-compromise navigation.
- 2021 · historyAI/ML Supply Chain Vulnerabilities
Research into AI supply chain attacks highlighted how adversaries could move laterally from compromised development environments into production model pipelines.
- 2023 · historyOWASP Top 10 for LLMs
The OWASP Top 10 for LLMs identified insecure plugin design and indirect prompt injection as vectors that facilitate lateral movement between AI agents and internal systems.
- 2024 · historyMITRE ATLAS Integration
The MITRE ATLAS framework specifically mapped lateral movement techniques to AI-integrated environments, focusing on how attackers pivot between model training servers and data repositories.