For the travel services objective to store and rotate third-party API credentials used by an AI agent, Resource-based policy addresses a different task—control which AWS principals can read a sensitive data bucket used for retrieval. Its operating mechanism differs from that of AWS Secrets Manager: a resource policy can specify allowed principals and access conditions; this does not satisfy the requested outcome.

For the travel services objective to store and rotate third-party API credentials used by an AI agent, Amazon Bedrock AgentCore Identity addresses a different task—manage agent authentication and delegated access to external resources. Its operating mechanism differs from that of AWS Secrets Manager: AgentCore Identity helps agents use managed identity and credentials for authorized connections; this does not satisfy the requested outcome.

For the travel services objective to store and rotate third-party API credentials used by an AI agent, Separation of duties addresses a different task—require a different employee to approve a high-risk AI production deployment. Its operating mechanism differs from that of AWS Secrets Manager: separate roles and approval responsibilities reduce unilateral unauthorized changes; this does not satisfy the requested outcome.

For the online retail objective to reconstruct which AI request and authorized tool call caused a disputed action, Vulnerability management addresses a different task—detect software flaws in infrastructure hosting an AI application. Its operating mechanism differs from that of Prompt and response audit trail: vulnerability assessments and patching reduce exposure in software dependencies; this does not satisfy the requested outcome.

For the online retail objective to reconstruct which AI request and authorized tool call caused a disputed action, Prompt injection defense addresses a different task—keep a retrieved page from issuing unauthorized instructions to an assistant. Its operating mechanism differs from that of Prompt and response audit trail: treat retrieved text as untrusted data and constrain tool actions with external checks; this does not satisfy the requested outcome.

For the online retail objective to reconstruct which AI request and authorized tool call caused a disputed action, Amazon CloudWatch monitoring addresses a different task—observe model-backed application errors, latency and operating metrics. Its operating mechanism differs from that of Prompt and response audit trail: CloudWatch collects metrics and operational logs for supported services and applications; this does not satisfy the requested outcome.

For the insurance operations objective to observe model-backed application errors, latency and operating metrics, Vulnerability management addresses a different task—detect software flaws in infrastructure hosting an AI application. Its operating mechanism differs from that of Amazon CloudWatch monitoring: vulnerability assessments and patching reduce exposure in software dependencies; this does not satisfy the requested outcome.

For the insurance operations objective to observe model-backed application errors, latency and operating metrics, Prompt and response audit trail addresses a different task—reconstruct which AI request and authorized tool call caused a disputed action. Its operating mechanism differs from that of Amazon CloudWatch monitoring: audit logs retain identity, time, operation and relevant trace information under privacy rules; this does not satisfy the requested outcome.

For the insurance operations objective to observe model-backed application errors, latency and operating metrics, Tool-action approval gate addresses a different task—require a person to confirm that an assistant may transfer funds or delete records. Its operating mechanism differs from that of Amazon CloudWatch monitoring: an approval checkpoint prevents high-impact tool invocations without explicit authorization; this does not satisfy the requested outcome.

For the insurance operations objective to observe model-backed application errors, latency and operating metrics, Prompt injection defense addresses a different task—keep a retrieved page from issuing unauthorized instructions to an assistant. Its operating mechanism differs from that of Amazon CloudWatch monitoring: treat retrieved text as untrusted data and constrain tool actions with external checks; this does not satisfy the requested outcome.

For the university services objective to block model-generated fields that break a permitted-data schema, Amazon Bedrock Guardrails addresses a different task—apply managed safety filters to inputs and outputs in a GenAI application. Its operating mechanism differs from that of Output validation: guardrails can help filter harmful content and sensitive information under configured policies; this does not satisfy the requested outcome.

For the university services objective to block model-generated fields that break a permitted-data schema, PII minimization addresses a different task—avoid sending unneeded customer identifiers to a third-party tool. Its operating mechanism differs from that of Output validation: data minimization removes unnecessary personal information before processing; this does not satisfy the requested outcome.

For the university services objective to block model-generated fields that break a permitted-data schema, Amazon Macie addresses a different task—discover potentially sensitive customer information stored in Amazon S3. Its operating mechanism differs from that of Output validation: Macie analyzes S3 data to identify sensitive information and related risks; this does not satisfy the requested outcome.

For the human-resources support objective to discover potentially sensitive customer information stored in Amazon S3, Data leakage prevention addresses a different task—prevent a helpdesk assistant from disclosing restricted customer records. Its operating mechanism differs from that of Amazon Macie: combine permission-scoped retrieval, output checking and monitored tool access; this does not satisfy the requested outcome.

For the human-resources support objective to discover potentially sensitive customer information stored in Amazon S3, Amazon Bedrock Guardrails addresses a different task—apply managed safety filters to inputs and outputs in a GenAI application. Its operating mechanism differs from that of Amazon Macie: guardrails can help filter harmful content and sensitive information under configured policies; this does not satisfy the requested outcome.

For the human-resources support objective to discover potentially sensitive customer information stored in Amazon S3, PII minimization addresses a different task—avoid sending unneeded customer identifiers to a third-party tool. Its operating mechanism differs from that of Amazon Macie: data minimization removes unnecessary personal information before processing; this does not satisfy the requested outcome.

For the online retail objective to segregate compute resources used by an internal AI system, AgentCore Policy addresses a different task—limit tool actions that an AI agent is allowed to perform. Its operating mechanism differs from that of Amazon VPC network isolation: AgentCore Policy supports policy-based enforcement of agent actions; this does not satisfy the requested outcome.

For the online retail objective to segregate compute resources used by an internal AI system, Prompt and response audit trail addresses a different task—reconstruct which AI request and authorized tool call caused a disputed action. Its operating mechanism differs from that of Amazon VPC network isolation: audit logs retain identity, time, operation and relevant trace information under privacy rules; this does not satisfy the requested outcome.

For the online retail objective to segregate compute resources used by an internal AI system, AWS PrivateLink addresses a different task—establish private connectivity from an application VPC to supported AWS services. Its operating mechanism differs from that of Amazon VPC network isolation: privateLink uses private endpoint connectivity without requiring public internet traversal; this does not satisfy the requested outcome.

For the media publishing objective to let a managed workload access AI-related AWS resources without embedded long-term keys, Resource-based policy addresses a different task—control which AWS principals can read a sensitive data bucket used for retrieval. Its operating mechanism differs from that of IAM role for workloads: a resource policy can specify allowed principals and access conditions; this does not satisfy the requested outcome.

For the media publishing objective to let a managed workload access AI-related AWS resources without embedded long-term keys, AgentCore Policy addresses a different task—limit tool actions that an AI agent is allowed to perform. Its operating mechanism differs from that of IAM role for workloads: AgentCore Policy supports policy-based enforcement of agent actions; this does not satisfy the requested outcome.

For the media publishing objective to let a managed workload access AI-related AWS resources without embedded long-term keys, IAM least privilege addresses a different task—limit an application's Bedrock model calls to only necessary AWS permissions. Its operating mechanism differs from that of IAM role for workloads: IAM permissions authorize specific service actions and resources for a principal; this does not satisfy the requested outcome.

For the university services objective to detect unexpected changes to model-training objects during transfer, Data cataloging addresses a different task—help data owners find the approved source and owner of each dataset. Its operating mechanism differs from that of Data checksum verification: a catalog records metadata, ownership and classification of data assets; this does not satisfy the requested outcome.

For the university services objective to detect unexpected changes to model-training objects during transfer, Dataset lineage addresses a different task—explain which source and transformation produced a feature used in evaluation. Its operating mechanism differs from that of Data checksum verification: lineage records the path from original data through processing to model use; this does not satisfy the requested outcome.

For the university services objective to detect unexpected changes to model-training objects during transfer, RAG factual grounding addresses a different task—reduce unsupported answers by providing approved source documents at inference. Its operating mechanism differs from that of Data checksum verification: grounding gives the model retrieved evidence relevant to the question; this does not satisfy the requested outcome.

For the university services objective to manage encryption keys protecting customer documents stored for an AI workflow, Amazon CloudWatch monitoring addresses a different task—observe model-backed application errors, latency and operating metrics. Its operating mechanism differs from that of AWS KMS encryption: CloudWatch collects metrics and operational logs for supported services and applications; this does not satisfy the requested outcome.

For the university services objective to manage encryption keys protecting customer documents stored for an AI workflow, Encryption at rest addresses a different task—protect stored embeddings and documents against exposure of underlying storage media. Its operating mechanism differs from that of AWS KMS encryption: at-rest encryption protects persisted data with managed cryptographic controls; this does not satisfy the requested outcome.

For the university services objective to manage encryption keys protecting customer documents stored for an AI workflow, Encryption in transit addresses a different task—protect a customer's prompt while it travels between client and service. Its operating mechanism differs from that of AWS KMS encryption: tLS encrypts network communications and helps prevent interception in transit; this does not satisfy the requested outcome.

For the travel services objective to limit tool actions that an AI agent is allowed to perform, AWS Secrets Manager addresses a different task—store and rotate third-party API credentials used by an AI agent. Its operating mechanism differs from that of AgentCore Policy: secrets Manager provides managed secret storage and rotation integration; this does not satisfy the requested outcome.

For the travel services objective to limit tool actions that an AI agent is allowed to perform, IAM least privilege addresses a different task—limit an application's Bedrock model calls to only necessary AWS permissions. Its operating mechanism differs from that of AgentCore Policy: IAM permissions authorize specific service actions and resources for a principal; this does not satisfy the requested outcome.

For the travel services objective to limit tool actions that an AI agent is allowed to perform, IAM role for workloads addresses a different task—let a managed workload access AI-related AWS resources without embedded long-term keys. Its operating mechanism differs from that of AgentCore Policy: an IAM role provides temporary credentials to trusted AWS service workloads; this does not satisfy the requested outcome.

For the travel services objective to limit tool actions that an AI agent is allowed to perform, Separation of duties addresses a different task—require a different employee to approve a high-risk AI production deployment. Its operating mechanism differs from that of AgentCore Policy: separate roles and approval responsibilities reduce unilateral unauthorized changes; this does not satisfy the requested outcome.

For the university services objective to help data owners find the approved source and owner of each dataset, RAG factual grounding addresses a different task—reduce unsupported answers by providing approved source documents at inference. Its operating mechanism differs from that of Data cataloging: grounding gives the model retrieved evidence relevant to the question; this does not satisfy the requested outcome.

For the university services objective to help data owners find the approved source and owner of each dataset, Data checksum verification addresses a different task—detect unexpected changes to model-training objects during transfer. Its operating mechanism differs from that of Data cataloging: checksums provide an integrity signal when expected hashes match the received object; this does not satisfy the requested outcome.

For the university services objective to help data owners find the approved source and owner of each dataset, Document provenance checks addresses a different task—confirm that evidence used in an answer comes from an approved current policy source. Its operating mechanism differs from that of Data cataloging: provenance metadata identifies author, revision and authorized origin of documents; this does not satisfy the requested outcome.