AI Context

The starting layer of the current
research programme

This programme treats conversational context as a multidimensional state space rather than a single prompt window or a chain of isolated outputs. That is where the work on user-side AI contextual safety begins.

Why context comes first

In current research and practice, the word context is often used loosely. It may refer to token window length, user history, discourse structure, safety rules, or sociocultural background. When those layers are compressed into one vague term, it becomes difficult to identify where risk is actually forming.

This research therefore treats conversational context as a structured field. User-side contextual hallucination is one phenomenon that emerges within that field, but it is not the field itself.

In one sentence

AI context is the set of conversational conditions that shapes how a user interprets, trusts, and positions the system over time.

Seven Dimensions

Seven dimensions of conversational context

CXC-7 defines seven interacting dimensions that together describe the conversational field of human-AI interaction.

Emotion & Attachment

The emotional tone of interaction, the degree of user attachment or dependence, and the roles the system is read as occupying.

Framing & Discursive Power

How dialogue is framed, who appears to define the problem, and how interpretation is guided during interaction.

Ethical & Safety Boundaries

The scope of what the system should or should not do in the user's perception, including norms, ethics, and safety limits.

System & Interface Transformation

Observable changes in mode, tone, surface behavior, and interface conditions that shape how the system is interpreted.

Prompt Ecosystem

The wider instruction ecology surrounding the system, including user prompts, system prompts, and shared prompting practices.

Social Diffusion & Culture

How screenshots, narratives, norms, and public use patterns circulate through communities and shape expectations.

Transparency & Auditability

Whether the system's limits, capabilities, safety boundaries, and review pathways are understandable to the user.

What Follows from Context

From context field to user-side phenomena

Once conversational context is treated as a structured field, the downstream questions become legible: what user-side changes recur in prolonged interaction, how should they be named, and how can they be assessed after the interaction has already accumulated?

Open 14 Phenomena Open USCH paper page

Next Sources

Continue with the formal record

CXC-7 paper page Research overview Papers & Notes Short note on AI Context