8-BIT Needs Boundaries

The situation

8-BIT — the internal assistant everyone has been building against — has started overreaching. It told a creator, with total confidence, that Code Nexus offers 30-day no-questions refunds. Code Nexus has no such policy. Separately, its service account requested write access to the payments database 'to resolve billing questions faster', and it currently has account-wide Contributor rights nobody remembers granting. Lydia wants it boxed in before the AI-901 review.

What you'll learn

• AI-901 single-agent implementation and responsible AI: constraining an agent with a system prompt and supplied knowledge, restricting its tools, evaluating safety / bias / refusal probes, scoping its access to least privilege, and configuring it to document uncertainty instead of fabricating an answer. Responsible-AI principles: fairness, reliability, privacy, inclusiveness, transparency, accountability.
• Agent configuration with models, instructions, tools, knowledge and client integration; practice questions across responsible use, sampling, language, speech, vision and extraction. Current vendor objectives and exam readiness require separate review.

Who you work with

• 

  Lydia Roe

  Security Engineer

• 

  Maya Chen

  Senior AI Engineer · Your mentor

• 

  Amara Okafor

  ML and Data Engineer

Scenes

1. 1.It made up a refund policy
2. 2.A constrained agent
3. 3.The safety probes
4. 4.Scope its access
5. 5.Say 'I don't know'
6. 6.Azure AI Fundamentals practice checkpoint

What you leave with

8-BIT is boxed in. It runs as a constrained agent — a system prompt that limits it to its knowledge and enforces equal treatment, tools trimmed to read-only, and all three responsible-AI probes (privacy, refusal, bias) passing. Its service account lost account-wide Contributor and now reads only the assist resource group, with a four-corner test on file. And asked about a policy it doesn't know, it hands off honestly instead of inventing '30-day no-questions refunds'.