AWS Bedrock Third-Party or External Knowledge Base Associated to Agent

ELASTIC detection rule by Elastic · Elastic License v2

Detects when an Amazon Bedrock agent is associated with, or updated to use, a knowledge base via the AssociateAgentKnowledgeBase, or UpdateAgentKnowledgeBase API actions. Bedrock agents consume knowledge base (RAG) content as trusted context for the model. By wiring an agent to an externally controlled or third-party knowledge base, or by swapping in an attacker-controlled knowledge base, an adversary can redraw the agent's trust boundary toward an untrusted source. This is a software-supply-chain compromise and an indirect prompt-injection delivery vector: poisoned or adversarial content served from the associated knowledge base is treated as authoritative by the agent. Validate that the associated knowledge base, and any underlying data source, is owned and controlled by your organization.

[metadata]
creation_date = "2026/06/04"
integration = ["aws"]
maturity = "production"
updated_date = "2026/06/04"

[rule]
author = ["Elastic"]
description = """
Detects when an Amazon Bedrock agent is associated with, or updated to use, a knowledge base via the AssociateAgentKnowledgeBase, 
or UpdateAgentKnowledgeBase API actions. Bedrock agents consume knowledge base (RAG) content as trusted context for the model. 
By wiring an agent to an externally controlled or third-party knowledge base, or by swapping in an attacker-controlled knowledge base, 
an adversary can redraw the agent's trust boundary toward an untrusted source. This is a software-supply-chain compromise and an indirect 
prompt-injection delivery vector: poisoned or adversarial content served from the associated knowledge base is treated as authoritative
by the agent. Validate that the associated knowledge base, and any underlying data source, is owned and controlled by your organization.
"""
false_positives = [
    """
    Legitimate platform, ML, or application teams may associate or update knowledge bases on Bedrock agents as part of
    normal development, onboarding, or RAG pipeline changes. Verify that the actor identity, user agent, and source IP
    correspond to expected automation or authorized engineers, and that the associated knowledge base is an approved,
    organization-owned resource. If known behavior is causing false positives, it can be exempted from the rule.
    """,
]
from = "now-6m"
index = ["logs-aws.cloudtrail-*"]
language = "kuery"
license = "Elastic License v2"
name = "AWS Bedrock Third-Party or External Knowledge Base Associated to Agent"
note = """## Triage and analysis

### Investigating AWS Bedrock Third-Party or External Knowledge Base Associated to Agent

Amazon Bedrock agents use knowledge bases to retrieve content that is injected into the model's context as
trusted, authoritative information (Retrieval-Augmented Generation). The `AssociateAgentKnowledgeBase`, and 
`UpdateAgentKnowledgeBase` actions change which knowledge base an agent trusts. Because the model consumes this 
content as ground truth, redirecting an agent toward an externally controlled or attacker-supplied knowledge base 
is a supply-chain and indirect prompt-injection delivery vector — distinct from poisoning the content of a knowledge 
base the agent already trusts.

#### Possible investigation steps

- **Identify the actor and context**
  - Review `aws.cloudtrail.user_identity.arn`, `aws.cloudtrail.user_identity.type`, `user_agent.original`, and `source.ip`.
  - Confirm a related change request exists (RAG pipeline change, agent onboarding, model improvement work).
- **Validate the association**
  - In `aws.cloudtrail.flattened.request_parameters`, identify the `agentId`, `knowledgeBaseId`, and any third-party
    or external endpoint/configuration referenced.
  - Confirm the knowledge base and its underlying data source are owned by your organization and not an external account.
- **Assess blast radius**
  - Determine which applications or users invoke the affected agent and what sensitivity of decisions it drives.
  - Check `aws.cloudtrail.flattened.response_elements` for the resulting association state.
- **Correlate activity**
  - Look for preceding enumeration (`ListAgents`, `ListKnowledgeBases`, `GetAgent`) or creation of new knowledge
    bases and data sources from the same identity.

### False positive analysis

- **Planned RAG changes**: ML/platform teams routinely associate or update knowledge bases. Validate via ticket
  and confirm the resource is an approved, organization-owned knowledge base.
- **Automation**: IaC or CI/CD pipelines may manage agent–knowledge base associations during deployment.

### Response and remediation

- If unauthorized, dissociate the knowledge base from the agent and restore the approved configuration.
- Review the associated knowledge base and its data source for attacker-controlled or external content; quarantine if suspect.
- Audit the actor's recent Bedrock and IAM activity and rotate credentials if compromise is suspected.
- Restrict `bedrock:AssociateAgentKnowledgeBase`, `bedrock:UpdateAgentKnowledgeBase`, and third-party association
  permissions to a small set of trusted roles.
"""
references = [
    "https://docs.aws.amazon.com/bedrock/latest/userguide/agents.html",
    "https://docs.aws.amazon.com/bedrock/latest/userguide/knowledge-base.html"
]
risk_score = 47
rule_id = "168b5538-4d54-4269-b4f5-786fa49dd850"
severity = "medium"
tags = [
    "Domain: Cloud",
    "Domain: LLM",
    "Data Source: Amazon Web Services",
    "Data Source: AWS",
    "Data Source: AWS CloudTrail",
    "Use Case: Threat Detection",
    "Resources: Investigation Guide",
    "Tactic: Persistence",
]
timestamp_override = "event.ingested"
type = "new_terms"

query = '''
data_stream.dataset: "aws.cloudtrail"
    and event.provider: "bedrock.amazonaws.com"
    and event.action: (
        "AssociateAgentKnowledgeBase" or
        "UpdateAgentKnowledgeBase"
    )
    and event.outcome: "success"
'''


[[rule.threat]]
framework = "MITRE ATT&CK"

[[rule.threat.technique]]
id = "T1505"
name = "Server Software Component"
reference = "https://attack.mitre.org/techniques/T1505/"

[rule.threat.tactic]
id = "TA0003"
name = "Persistence"
reference = "https://attack.mitre.org/tactics/TA0003/"

[rule.investigation_fields]
field_names = [
    "@timestamp",
    "user.name",
    "user_agent.original",
    "source.ip",
    "aws.cloudtrail.user_identity.arn",
    "aws.cloudtrail.user_identity.type",
    "aws.cloudtrail.user_identity.access_key_id",
    "event.action",
    "event.provider",
    "event.outcome",
    "cloud.account.id",
    "cloud.region",
    "aws.cloudtrail.flattened.request_parameters",
    "aws.cloudtrail.flattened.response_elements",
]
 
[rule.new_terms]
field = "new_terms_fields"
value = ["aws.cloudtrail.user_identity.arn"]

[[rule.new_terms.history_window_start]]
field = "history_window_start"
value = "now-7d"

[[ currentRule.title && currentRule.title.length > 60 ? currentRule.title.slice(0, 60) + '…' : currentRule.title ]]

85cea6b3-e225-4e18-8305-204658d20433

Propose edits, view related rules, and engage with the community through comments.

Executable code, not a declarative rule

This submission is arbitrary [[ currentRule.format ? currentRule.format.toUpperCase() : '' ]] code, not a declarative detection rule — review it as you would any script before running it, and note that it cannot be safely aggregated or auto-executed alongside YARA/Sigma/Suricata/CRS rules.

Rejected pattern present

[[ reason ]]

What does this mean?

Format & documentation suggestions

Not a security issue — this rule is valid and was accepted as-is. These are optional fields/conventions its own format recommends, mainly affecting its quality score and how well it documents itself for other reviewers.

[[ reason ]]

What do these messages mean?

[[ currentRule.title && currentRule.title.length > 60 ? currentRule.title.slice(0, 60) + '…' : currentRule.title ]]

[[ currentRule.title ]]

Community Discussion