Question 530 A CI pipeline upgrades prompt text but leaves the registered model artifact unchanged and provides no way to identify the deployed combination. What prevention measure is most targeted to this underlying cause? Require release metadata linking code, prompt, model and index versions for each environment. Use immutable prompt versions, approval records and controlled release aliases. Use parameterized environment targets, peer review and automated resource-diff checks. Make environment-specific least-privilege checks a required CI promotion gate. ExplanationCorrect Answer A:Generation depends on multiple independently mutable assets; unversioned prompt drift prevents reproducible rollbacks. Other Answers B:Generation depends on multiple independently mutable assets; unversioned prompt drift prevents reproducible rollbacks. C:This control addresses a different exposure: Developers manually copy an agent notebook and vector-search configuration into production, accidentally pointing the prod app at the dev catalog. It would not prevent this failure, because generation depends on multiple independently mutable assets; unversioned prompt drift prevents reproducible rollbacks. D:That generation depends on multiple independently mutable assets; unversioned prompt drift prevents reproducible rollbacks.
Question 531 A Git workflow deploys an agent without running tests against the target workspace's security permissions, causing launch-time authorization failures. Which implementation decision best restores the intended behavior? Version prompt, model artifact, retrieval configuration and bundle commit together as one promotion unit. Configure the production agent identity and Unity Catalog permissions for the managed data tool, then test the exact delegated query. Define environment-specific catalog, model endpoint and index bindings in a reviewed Declarative Automation Bundle deployment. Add staging integration tests with the same scoped service-principal grants before production promotion. ExplanationCorrect Answer D:The action targets the stated cause: Local unit tests cannot establish real workspace permissions or endpoint accessibility under deployment identity. Check the repair by: execute allowed and forbidden model/index access tests in staging and verify the expected audit identities. Other Answers A:Local unit tests cannot establish real workspace permissions or endpoint accessibility under deployment identity. B:Local unit tests cannot establish real workspace permissions or endpoint accessibility under deployment identity. C:Local unit tests cannot establish real workspace permissions or endpoint accessibility under deployment identity.
Question 532 A Git workflow deploys an agent without running tests against the target workspace's security permissions, causing launch-time authorization failures. Which underlying technical principle explains why this occurs? Local unit tests cannot establish real workspace permissions or endpoint accessibility under deployment identity. Manual configuration drift can silently redirect production to incorrect data and model resources across environments. Generation depends on multiple independently mutable assets; unversioned prompt drift prevents reproducible rollbacks. Managed MCP discovery does not automatically grant rights to governed tables or semantic resources. ExplanationCorrect Answer A:Add staging integration tests with the same scoped service-principal grants before production promotion. Other Answers B:Local unit tests cannot establish real workspace permissions or endpoint accessibility under deployment identity. C:That mechanism describes another problem: A CI pipeline upgrades prompt text but leaves the registered model artifact unchanged and provides no way to identify the deployed combination. It does not explain this case, where local unit tests cannot establish real workspace permissions or endpoint accessibility under deployment identity. D:Local unit tests cannot establish real workspace permissions or endpoint accessibility under deployment identity.
Question 533 An agent is granted unrestricted access to an external MCP connector, including a destructive remove-record tool, although its task only reads incident metadata. What is the most targeted corrective step for this observed failure? Store only approved non-sensitive conversation state and keep secrets in managed credential storage with short-lived tool access. Register and govern the connector with an allowlisted read-only tool subset and a scoped execution identity. Configure the production agent identity and Unity Catalog permissions for the managed data tool, then test the exact delegated query. Use approved server-side credentials and tool authentication with scoped network access, never model-visible plaintext secrets. ExplanationCorrect Answer B:The action targets the stated cause: MCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements. Check the repair by: enumerate visible tools as the agent and verify the destructive operation cannot be invoked. Other Answers A:MCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements. C:MCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements. D:MCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements.
Question 534 An agent is granted unrestricted access to an external MCP connector, including a destructive remove-record tool, although its task only reads incident metadata. What is the decisive mechanism behind the observed result? Persisted memory expands the exposure and retention of sensitive data; tool authentication is separate from session context. MCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements. MCP tool descriptions and model prompts are not secret-management boundaries; credentials belong in the service runtime. Managed MCP discovery does not automatically grant rights to governed tables or semantic resources. ExplanationCorrect Answer B:Register and govern the connector with an allowlisted read-only tool subset and a scoped execution identity. Other Answers A:MCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements. C:MCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements. D:That mechanism describes another problem: A Databricks-managed Genie MCP tool works in development, but a production custom agent cannot call it because the workspace identity lacks required data grants. It does not explain this case, where mCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements.
Question 535 An agent is granted unrestricted access to an external MCP connector, including a destructive remove-record tool, although its task only reads incident metadata. Which verification is strongest for this particular failure mode? Inspect sampled tool traces and outbound requests to verify no API secrets appear in model context. Enumerate visible tools as the agent and verify the destructive operation cannot be invoked. Inspect stored state after a simulated conversation and confirm no token or protected record remains. Run the tool with the production principal and inspect both the permission denial before grants and authorized query after. ExplanationCorrect Answer B:MCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements. Other Answers A:Enumerate visible tools as the agent and verify the destructive operation cannot be invoked. C:Enumerate visible tools as the agent and verify the destructive operation cannot be invoked. D:Enumerate visible tools as the agent and verify the destructive operation cannot be invoked.
Question 536 An agent is granted unrestricted access to an external MCP connector, including a destructive remove-record tool, although its task only reads incident metadata. What change to engineering governance would reduce this risk over time? Apply field-level retention rules, encryption and memory redaction tests. Review MCP tool allowlists and credentials whenever a connector version changes. Require secret scanning, short-lived credentials and egress controls for custom MCP services. Maintain principal-to-tool-to-data entitlement maps for each managed MCP integration. ExplanationCorrect Answer B:MCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements. Other Answers A:MCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements. C:This control addresses a different exposure: An engineering team hosts a custom MCP server in a Databricks App but passes API keys through plain prompt text rather than through authenticated tool transport. It would not prevent this failure, because mCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements. D:That mCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements.
Question 537 A Databricks-managed Genie MCP tool works in development, but a production custom agent cannot call it because the workspace identity lacks required data grants. Which engineering action most directly addresses this specific requirement? Add staging integration tests with the same scoped service-principal grants before production promotion. Register and govern the connector with an allowlisted read-only tool subset and a scoped execution identity. Configure the production agent identity and Unity Catalog permissions for the managed data tool, then test the exact delegated query. Use approved server-side credentials and tool authentication with scoped network access, never model-visible plaintext secrets. ExplanationCorrect Answer C:The action targets the stated cause: Managed MCP discovery does not automatically grant rights to governed tables or semantic resources. Check the repair by: run the tool with the production principal and inspect both the permission denial before grants and authorized query after. Other Answers A:Managed MCP discovery does not automatically grant rights to governed tables or semantic resources. B:Managed MCP discovery does not automatically grant rights to governed tables or semantic resources. D:Managed MCP discovery does not automatically grant rights to governed tables or semantic resources.
Question 538 A Databricks-managed Genie MCP tool works in development, but a production custom agent cannot call it because the workspace identity lacks required data grants. What architectural constraint is most directly responsible? Managed MCP discovery does not automatically grant rights to governed tables or semantic resources. Local unit tests cannot establish real workspace permissions or endpoint accessibility under deployment identity. MCP tool descriptions and model prompts are not secret-management boundaries; credentials belong in the service runtime. MCP provides a tool interface; access policies, authentication and individual tool permissions remain separate requirements. ExplanationCorrect Answer A:Configure the production agent identity and Unity Catalog permissions for the managed data tool, then test the exact delegated query. Other Answers B:Managed MCP discovery does not automatically grant rights to governed tables or semantic resources. C:That mechanism describes another problem: An engineering team hosts a custom MCP server in a Databricks App but passes API keys through plain prompt text rather than through authenticated tool transport. It does not explain this case, where managed MCP discovery does not automatically grant rights to governed tables or semantic resources. D:Managed MCP discovery does not automatically grant rights to governed tables or semantic resources.
Question 539 A Databricks-managed Genie MCP tool works in development, but a production custom agent cannot call it because the workspace identity lacks required data grants. What validation evidence would most specifically confirm the resolution? Enumerate visible tools as the agent and verify the destructive operation cannot be invoked. Run the tool with the production principal and inspect both the permission denial before grants and authorized query after. Execute allowed and forbidden model/index access tests in staging and verify the expected audit identities. Inspect sampled tool traces and outbound requests to verify no API secrets appear in model context. ExplanationCorrect Answer B:Managed MCP discovery does not automatically grant rights to governed tables or semantic resources. Other Answers A:Run the tool with the production principal and inspect both the permission denial before grants and authorized query after. C:Run the tool with the production principal and inspect both the permission denial before grants and authorized query after. D:Run the tool with the production principal and inspect both the permission denial before grants and authorized query after.
Question 540 A Databricks-managed Genie MCP tool works in development, but a production custom agent cannot call it because the workspace identity lacks required data grants. What prevention measure is most targeted to this underlying cause? Make environment-specific least-privilege checks a required CI promotion gate. Require secret scanning, short-lived credentials and egress controls for custom MCP services. Maintain principal-to-tool-to-data entitlement maps for each managed MCP integration. Review MCP tool allowlists and credentials whenever a connector version changes. ExplanationCorrect Answer C:Managed MCP discovery does not automatically grant rights to governed tables or semantic resources. Other Answers A:This control addresses a different exposure: A Git workflow deploys an agent without running tests against the target workspace's security permissions, causing launch-time authorization failures. It would not prevent this failure, because managed MCP discovery does not automatically grant rights to governed tables or semantic resources. B:That managed MCP discovery does not automatically grant rights to governed tables or semantic resources. D:Managed MCP discovery does not automatically grant rights to governed tables or semantic resources.
