Account-scoped context is private session data bound to a specific user account, such as chat history, saved preferences, subscription tier, identity details, or other personalization state. In AI systems, this context is often available to the model or its tools so responses can feel continuous and personalized.
It matters because any untrusted input that can influence an assistant may try to leak or misuse that private state. In prompt-injection attacks, malicious web content can be crafted to trick a browser-enabled AI into revealing account-scoped context or into sending it to an attacker through a tool call, URL, or other outbound request. Defenses include strict separation between fetched content and private session data, narrow tool permissions, provenance checks, and output filtering that prevents sensitive context from leaving the trusted boundary.