liveBook · Manning

Data access

As explained in AI Governance

The ability of a system to retrieve or otherwise access information.

Data access lets a GenAI application retrieve sensitive information at runtime. Common connections include databases, CRMs, document stores, and RAG pipelines. section 4.2

Without runtime data access, manipulated input may produce bad text but has limited impact. With access to live systems, prompt injection can expose sensitive data. A successful leak also requires external communication, such as fetching a URL or rendering an image link. section 4.2 section 4.6

A potential data leak

An application with untrusted input, access to customer records, and a channel for external communication has all three elements of the Lethal Trifecta. Breaking any one element prevents the described attack. section 4.6

Figure 4.2 Security Risk Factors. Each additional factor compounds the risk.

Use the diagram to see data access as one factor in a security risk ladder where additional capabilities compound risk. section 4.2

How AI Governance covers it

How deep it goes
Defines runtime retrieval, names connected systems, and contrasts contained bad output with breach exposure without detailing access controls.
Examples
Names customer records, internal policies, trade secrets, source code, financial data, databases, CRMs, document stores, and RAG pipelines.
What you can do after
Evaluate whether runtime retrieval could turn prompt injection into a potential data breach.
How it connects
Links data access with prompt injection, RAG, sensitive data, and the Lethal Trifecta's external-communication channel.

Also explained in

Compare how 2 books explain data access →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.