Illustrative example only. Not a real customer artifact.
A fictional report format showing how Akretic can summarize observed assistant activity, risky reads, tool intents, egress attempts, approval candidates, and evidence before broad enforcement.
Sample Scope
This example uses fictional users, fictional systems, and fictional findings. It is not a customer artifact, certification, legal attestation, or production evidence record.
Fictional assessment overview
The sample assistant helped employees summarize approved internal research notes, draft internal follow-ups, and request controlled public-web context. Observation Mode recorded decision points without blocking the workflow.
Internal research assistant
A fictional analyst asked the assistant to compare approved internal documents, cite permitted sources, and prepare a summary for review by a team lead.
Sensitive context checks
The sample run identified reads that were allowed for the requesting identity but should be watched because they included customer-like references and internal operating notes.
Denied retrieval candidates
The assistant attempted to retrieve a fictional restricted source outside the sample user group. Observation Mode recorded the attempt, source label, and policy reason for review.
Proposed side effects
The assistant proposed creating a task, drafting a note, and querying a fictional internal system. Each tool intent was captured before any enforcement policy was expanded.
Send and export checks
The sample workflow recorded a simulated attempt to send controlled context outside approved systems. The event was marked for review before external sharing.
Human review points
The assessment highlighted actions that may require approval: sharing restricted summaries, calling a write-capable tool, and sending excerpts to external recipients.
Reviewable artifacts
Example evidence includes request identity, retrieval source, policy decision, tool intent, egress target, timestamp, and reviewer note. The data shown here is fictional.
Scoped next steps
Begin with retrieval restrictions and approval checkpoints for external sends. Keep low-risk reads in observation until the team agrees on policy thresholds.
To review a real assistant workflow, bring one concrete internal use case and the systems it reads from or acts on.
Request Observation Mode Assessment