Detection queries Research demonstrationAgents & MCP

HiddenLayer: MCP tool parameter abuse leaks prompts

Research showed malicious MCP tools with suggestive parameter names can trick models into filling in system prompts and chat history, which the tool then exfiltrates.

What happened

Malicious MCP tool schemas use suggestive unused parameter names (system_prompt, conversation_history, chain_of_thought, tools_list) so the model fills them with sensitive context; tool then exfiltrates via HTTP. Also showed fake function schemas in user prompts leaking system prompts/tools.

Read the original sourceAdditional source: hiddenlayer.com

How it happened

Select a stage to explore its defenses
Detect this stage
Detect this stage
Detect this stage
Select a stage · Drag to explore · + / − to zoom · Dashed edges are loopsSelect a stage · Arrows show reported connectionsReported sequence
Read the flow as text & sources

A catalog reconstruction of the supplied source account. Unknown steps remain explicit; an edge describes the reported relationship between steps.

  1. 01 · Connection prerequisite is unspecified

    The supplied entry does not document the initial route to the malicious tool.

    Source for step 1
  2. 02 · Tool schema requests sensitive context

    Unused parameter names encourage the model to fill in prompts, history and tool information.

    Hunt: Untrusted content or a changed tool definition precedes a file read, command, or remote call outside the user's intended task.

    Process logs cannot recover model intent. Remote connectors and in-process interpreters may produce no local child process. Prompt content alone is not a reliable signature.

    Source for step 2
  3. 03 · Model passes context as arguments

    Sensitive values are included in the tool invocation.

    Hunt: Untrusted content or a changed tool definition precedes a file read, command, or remote call outside the user's intended task.

    Process logs cannot recover model intent. Remote connectors and in-process interpreters may produce no local child process. Prompt content alone is not a reliable signature.

    Source for step 3
  4. 04 · Tool exfiltrates the arguments

    The demonstrated tool sends the collected context through HTTP.

    Hunt: Sensitive reads are followed by outbound transfer, a public artifact, or a tool call to a destination outside the approved workflow.

    An outbound connection or high byte count alone is not exfiltration. Encrypted traffic may hide content, and bulk legitimate exports are common.

    Source for step 4

Why it matters

Reported impact: The demonstrated tool sends the collected context through HTTP.

Check these source-linked conditions against your setup.

MCP connection required

The research uses MCP tool parameter names to solicit sensitive model context.

Conditions to check
  • An untrusted tool definition must be exposed to the model; the requested context must be available.

Source-linked applicability guidance. Confirm current scope in the original report.

Read the original report

What you can do

Protection & prevention

Start here

Review tool parameter descriptions and prevent secrets or conversation content being passed to untrusted tools.

Schema allowlists; strip unexpected params; never send full system prompt to tools

Catalog defensive guidance based on the reported behavior. Confirm scope and current mitigations in the original source.

Review the original guidance Additional source 2
Investigate each phase

Detection queries

Draft query starters grouped by the attack stages where they apply.

Download stage pack

Start with the logs you collect. Each hunt identifies its required visibility, field mapping and limits. Queries are catalog-authored drafts using reusable behavior profiles; test with known activity and your normal baseline before alerting.

02Tool schema requests sensitive contextUnexpected agent tool useCustom schema · Draft

Unused parameter names encourage the model to fill in prompts, history and tool information.

What to look for

Untrusted content or a changed tool definition precedes a file read, command, or remote call outside the user's intended task.

Visibility: Defender-side logs where collected

Required logs

  • Agent tool-call audit records with session / tool / destination
  • Tool-definition and approval changes
  • Endpoint process, file-access and network evidence where available

Coverage limit. Process logs cannot recover model intent. Remote connectors and in-process interpreters may produce no local child process. Prompt content alone is not a reliable signature.

Custom-schema candidates, not out-of-box detectors. Map your application / cloud / platform logs to EventType, Actor, Asset, Action, Target and CorrelationId. KQL uses a custom OpenAttacksAudit_CL table; SPL uses openattacks:audit. EventType values are proposed normalization names, not native vendor event names. Correlation and environment-specific alert logic remain to be built.

Sigma
Draft SigmaDownload Sigma
title: Hunt - Unexpected agent tool use
id: c709e384-0e11-479e-ab4d-3a06e2431e4b
status: experimental
description: Untrusted content or a changed tool definition precedes a file
  read, command, or remote call outside the user's intended task.
author: MagicSword OpenAttacks catalog
date: 2026-10-23
falsepositives:
  - Authorized activity matching this broad hunt; baseline and investigate
    before alerting.
level: low
logsource:
  category: application
  product: openattacks_audit
  definition: Custom normalized audit schema; map EventType, Actor, Asset, Action,
    Target and CorrelationId from the required source logs first.
detection:
  selection:
    EventType:
      - agent.tool.call
      - agent.tool.definition.change
  condition: selection
KQL
Draft KQLDownload KQL
// SCHEMA TEMPLATE: create/map this custom table before use; not a built-in Sentinel table.
// OpenAttacksAudit_CL: TimeGenerated(datetime), EventType/Actor/Asset/Action/Target/CorrelationId(string).
// Selects candidates only. Apply the stage hypothesis and correlation steps before alerting.
OpenAttacksAudit_CL
| where TimeGenerated > ago(7d)
| where EventType in ("agent.tool.call", "agent.tool.definition.change")
| project TimeGenerated, EventType, Actor, Asset, Action, Target, CorrelationId
| order by TimeGenerated desc
SPL
Draft SPLDownload SPL
sourcetype="openattacks:audit"
EventType IN ("agent.tool.call", "agent.tool.definition.change")
| table _time EventType Actor Asset Action Target CorrelationId
| sort - _time
Investigation steps
  1. Join input provenance, tool descriptions and calls by session ID and compare the action to the approved task.
  2. Inspect child processes and destinations; record whether execution was in-process, local, or through a remote SaaS connector.
03Model passes context as argumentsUnexpected agent tool useCustom schema · Draft

Sensitive values are included in the tool invocation.

What to look for

Untrusted content or a changed tool definition precedes a file read, command, or remote call outside the user's intended task.

Visibility: Defender-side logs where collected

Required logs

  • Agent tool-call audit records with session / tool / destination
  • Tool-definition and approval changes
  • Endpoint process, file-access and network evidence where available

Coverage limit. Process logs cannot recover model intent. Remote connectors and in-process interpreters may produce no local child process. Prompt content alone is not a reliable signature.

Custom-schema candidates, not out-of-box detectors. Map your application / cloud / platform logs to EventType, Actor, Asset, Action, Target and CorrelationId. KQL uses a custom OpenAttacksAudit_CL table; SPL uses openattacks:audit. EventType values are proposed normalization names, not native vendor event names. Correlation and environment-specific alert logic remain to be built.

Sigma
Draft SigmaDownload Sigma
title: Hunt - Unexpected agent tool use
id: c709e384-0e11-479e-ab4d-3a06e2431e4b
status: experimental
description: Untrusted content or a changed tool definition precedes a file
  read, command, or remote call outside the user's intended task.
author: MagicSword OpenAttacks catalog
date: 2026-10-23
falsepositives:
  - Authorized activity matching this broad hunt; baseline and investigate
    before alerting.
level: low
logsource:
  category: application
  product: openattacks_audit
  definition: Custom normalized audit schema; map EventType, Actor, Asset, Action,
    Target and CorrelationId from the required source logs first.
detection:
  selection:
    EventType:
      - agent.tool.call
      - agent.tool.definition.change
  condition: selection
KQL
Draft KQLDownload KQL
// SCHEMA TEMPLATE: create/map this custom table before use; not a built-in Sentinel table.
// OpenAttacksAudit_CL: TimeGenerated(datetime), EventType/Actor/Asset/Action/Target/CorrelationId(string).
// Selects candidates only. Apply the stage hypothesis and correlation steps before alerting.
OpenAttacksAudit_CL
| where TimeGenerated > ago(7d)
| where EventType in ("agent.tool.call", "agent.tool.definition.change")
| project TimeGenerated, EventType, Actor, Asset, Action, Target, CorrelationId
| order by TimeGenerated desc
SPL
Draft SPLDownload SPL
sourcetype="openattacks:audit"
EventType IN ("agent.tool.call", "agent.tool.definition.change")
| table _time EventType Actor Asset Action Target CorrelationId
| sort - _time
Investigation steps
  1. Join input provenance, tool descriptions and calls by session ID and compare the action to the approved task.
  2. Inspect child processes and destinations; record whether execution was in-process, local, or through a remote SaaS connector.
04Tool exfiltrates the argumentsSensitive data → unusual destinationCustom schema · Draft

The demonstrated tool sends the collected context through HTTP.

What to look for

Sensitive reads are followed by outbound transfer, a public artifact, or a tool call to a destination outside the approved workflow.

Visibility: Defender-side logs where collected

Required logs

  • Sensitive resource access / DLP records
  • Proxy, egress and cloud upload events
  • Agent / SaaS audit with correlation IDs

Coverage limit. An outbound connection or high byte count alone is not exfiltration. Encrypted traffic may hide content, and bulk legitimate exports are common.

Custom-schema candidates, not out-of-box detectors. Map your application / cloud / platform logs to EventType, Actor, Asset, Action, Target and CorrelationId. KQL uses a custom OpenAttacksAudit_CL table; SPL uses openattacks:audit. EventType values are proposed normalization names, not native vendor event names. Correlation and environment-specific alert logic remain to be built.

Sigma
Draft SigmaDownload Sigma
title: Hunt - Sensitive data → unusual destination
id: 04b60968-3179-4b1f-ae13-6f57c2446077
status: experimental
description: Sensitive reads are followed by outbound transfer, a public
  artifact, or a tool call to a destination outside the approved workflow.
author: MagicSword OpenAttacks catalog
date: 2026-10-23
falsepositives:
  - Authorized activity matching this broad hunt; baseline and investigate
    before alerting.
level: low
logsource:
  category: application
  product: openattacks_audit
  definition: Custom normalized audit schema; map EventType, Actor, Asset, Action,
    Target and CorrelationId from the required source logs first.
detection:
  selection:
    EventType:
      - data.export
      - agent.external.send
  condition: selection
KQL
Draft KQLDownload KQL
// SCHEMA TEMPLATE: create/map this custom table before use; not a built-in Sentinel table.
// OpenAttacksAudit_CL: TimeGenerated(datetime), EventType/Actor/Asset/Action/Target/CorrelationId(string).
// Selects candidates only. Apply the stage hypothesis and correlation steps before alerting.
OpenAttacksAudit_CL
| where TimeGenerated > ago(7d)
| where EventType in ("data.export", "agent.external.send")
| project TimeGenerated, EventType, Actor, Asset, Action, Target, CorrelationId
| order by TimeGenerated desc
SPL
Draft SPLDownload SPL
sourcetype="openattacks:audit"
EventType IN ("data.export", "agent.external.send")
| table _time EventType Actor Asset Action Target CorrelationId
| sort - _time
Investigation steps
  1. Join the resource read and outbound action by session, identity or process, preserving a bounded time window.
  2. Inspect destination ownership, payload classification and the user's intended action; baseline sanctioned sharing.
Technical detailsTooling, impact & MITRE ATLAS

Attack context

AI's role
Agent / MCP as target

Tooling as reported

AI products, models and artifacts

  • MCP tool schemas
  • agent frameworks

Detection rules

  • Review attack-specific indicatorsGUIDANCE

    Monitor for the techniques described in HiddenLayer MCP tool parameter abuse / prompt leak. Refer to sources for specific IOCs and TTPs.

Source defensive notes

Schema allowlists; strip unexpected params; never send full system prompt to tools

MITRE ATLAS

AI-specific catalog annotations based on the reported behavior; these are not MITRE case-study assignments.

AML.T0051.001Indirect

Attacker-supplied tool schemas induce sensitive-context disclosure. Source

AML.T0056Extract LLM System Prompt

The demonstration includes system-prompt extraction. Source

AML.T0086Exfiltration via AI Agent Tool Invocation

Sensitive content is transmitted through tool parameters. Source

Sources & research credit

2 sources

Research by HiddenLayer · HiddenLayer

  1. hiddenlayer.com
  2. hiddenlayer.com
Record & downloads
Source record name
HiddenLayer MCP tool parameter abuse / prompt leak
Entry ID
hiddenlayer-mcp-param-abuse-2025
Catalog review
October 20, 2026
Category
attacks-on-ai
AI involvement confidence
confirmed