Detection queries Reported campaignCloud & SaaS

LLMjacking via stolen cloud credentials (Sysdig)

Attackers used stolen cloud credentials to probe ten AI services and resell LLM access through proxies, leaving victims to pay the inference bills.

What happened

After a Laravel CVE-2021-3129 foothold, attackers stole cloud credentials and probed ten AI services (AI21, Anthropic, AWS Bedrock, Azure, ElevenLabs, MakerSuite, Mistral, OpenAI, OpenRouter, and GCP Vertex) to resell LLM access via reverse proxies while victims paid inference bills. Later Sysdig reporting showed 85,000+ Bedrock requests with Converse API abuse within 30 days of launch. Pillar Security separately reported the Bizarre Bazaar campaign targeting AI startups for unauthorized inference resale.

Read the original sourceAdditional source: Sysdig: Growing Dangers of LLMjackingAdditional source: Pillar Security: Bizarre Bazaar - LLMjacking Campaign Targeting AI Startups

How it happened

Select a stage to explore its defenses
Detect this stage
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 · Vulnerable Laravel provides a foothold

    The report identifies CVE-2021-3129 before credential theft.

    Hunt: A PHP or web-server worker launches a shell, interpreter, or reconnaissance utility outside its normal baseline.

    Legitimate CGI, maintenance, and deployment scripts can match. The query is a process-tree starter, not a complete web-shell detector.

    Source for step 1
  2. 02 · Cloud credentials are stolen

    Attackers obtain keys and probe multiple AI providers.

    Hunt: An unexpected process or tool accesses credential stores and is followed by token use or outbound transfer inconsistent with the owner.

    Defender DeviceFileEvents is not a general file-read audit trail. Collect suitable object-access, EDR or application logs; legitimate developer tools also read secrets.

    Source for step 2
  3. 03 · Inference access is proxied

    Authorized model access is resold through attacker reverse proxies.

    Hunt: A key enumerates Bedrock models from an unfamiliar origin, followed by unexpected model access or usage.

    The starter queries cover discovery only. Do not assume default CloudTrail management logs contain InvokeModel / Converse requests or prompt content.

    Source for step 3
  4. 04 · Victim accounts incur inference usage

    The report describes high-volume requests and potentially substantial costs.

    Hunt: A service key or role is used from an unfamiliar origin, gains broader access or causes inference usage outside its established baseline.

    Cloud control-plane logs alone may not include model prompts or inference calls. No endpoint policy is claimed to stop cloud-only key misuse.

    Source for step 4

Why it matters

Reported entry point: Vulnerable Laravel (CVE-2021-3129) leading to cloud credential theft. Reported impact: The report describes high-volume requests and potentially substantial costs.

Check these source-linked conditions against your setup.

AI API / cloud credentials

The reported path uses stolen AI-service or cloud credentials to resell model access.

Conditions to check
  • Confirm the affected service and credential type in the source; consumer chat use alone is not the reported condition.

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

Read the original report

What you can do

Protection & prevention

Start here

Inventory AI service keys, scope permissions, rotate exposed tokens and investigate anomalous usage.

Monitor Bedrock ListFoundationModels+Invoke/Converse from long-term AKIA keys; rotate exposed keys; SCP restrict Bedrock

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

Review the original guidance Additional source 2 Additional source 3
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.

01Vulnerable Laravel provides a footholdLinux web worker → shell / interpreterNative logs · Draft

The report identifies CVE-2021-3129 before credential theft.

What to look for

A PHP or web-server worker launches a shell, interpreter, or reconnaissance utility outside its normal baseline.

Visibility: Defender-side logs where collected

Required logs

  • Linux process creation (EDR / auditd normalized to process fields)
  • PHP / web access logs
  • File changes to web roots and PHP configuration

Coverage limit. Legitimate CGI, maintenance, and deployment scripts can match. The query is a process-tree starter, not a complete web-shell detector.

linux process candidates. Sigma needs normalized process events; KQL needs Defender coverage for that OS; SPL uses a custom openattacks:process sourcetype with Sysmon-style Image / ParentImage / CommandLine fields. Map raw auditd or Endpoint Security data first.

Sigma
Draft SigmaDownload Sigma
title: Hunt - Linux web worker → shell / interpreter
id: 068e005a-876e-4ba3-a414-791f9a1c2330
status: experimental
description: A PHP or web-server worker launches a shell, interpreter, or
  reconnaissance utility outside its normal baseline.
author: MagicSword OpenAttacks catalog
date: 2026-10-23
falsepositives:
  - Authorized activity matching this broad hunt; baseline and investigate
    before alerting.
level: low
logsource:
  category: process_creation
  product: linux
detection:
  selection:
    Image|endswith:
      - /sh
      - /bash
      - /python
      - /python3
      - /curl
      - /wget
    ParentImage|endswith:
      - /php-fpm
      - /php-fpm8.2
      - /apache2
      - /httpd
      - /nginx
  condition: selection
KQL
Draft KQLDownload KQL
// Draft hunt; requires Defender for Endpoint process telemetry for this OS.
DeviceProcessEvents
| where Timestamp > ago(7d)
| where FileName in~ ("sh", "bash", "python", "python3", "curl", "wget")
| where InitiatingProcessFileName in~ ("php-fpm", "php-fpm8.2", "apache2", "httpd", "nginx")
| project Timestamp, DeviceName, AccountName, FileName, ProcessCommandLine, InitiatingProcessFileName, InitiatingProcessCommandLine, SHA1
| order by Timestamp desc
SPL
Draft SPLDownload SPL
sourcetype="openattacks:process"
Image IN ("*/sh", "*/bash", "*/python", "*/python3", "*/curl", "*/wget")
ParentImage IN ("*/php-fpm", "*/php-fpm8.2", "*/apache2", "*/httpd", "*/nginx")
| table _time host User Image CommandLine ParentImage ParentCommandLine
| sort - _time
Investigation steps
  1. Trace the worker's children and correlate them with inbound requests.
  2. Review auto_prepend_file, .bashrc changes, /dev/shm execution, and outbound tunnels where the report describes them.
02Cloud credentials are stolenSensitive store access → credential useCustom schema · Draft

Attackers obtain keys and probe multiple AI providers.

What to look for

An unexpected process or tool accesses credential stores and is followed by token use or outbound transfer inconsistent with the owner.

Visibility: Defender-side logs where collected

Required logs

  • File-read / API access telemetry for relevant secret stores
  • Process identity and command lines
  • Identity, token and outbound-network audit logs

Coverage limit. Defender DeviceFileEvents is not a general file-read audit trail. Collect suitable object-access, EDR or application logs; legitimate developer tools also read secrets.

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 store access → credential use
id: 46c46c62-2d83-4198-a8bf-d6637094527b
status: experimental
description: An unexpected process or tool accesses credential stores and is
  followed by token use or outbound transfer inconsistent with the owner.
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:
      - secret.read
      - credential.store.read
  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 ("secret.read", "credential.store.read")
| project TimeGenerated, EventType, Actor, Asset, Action, Target, CorrelationId
| order by TimeGenerated desc
SPL
Draft SPLDownload SPL
sourcetype="openattacks:audit"
EventType IN ("secret.read", "credential.store.read")
| table _time EventType Actor Asset Action Target CorrelationId
| sort - _time
Investigation steps
  1. Identify the actor, process and exact store touched, then pivot to token use from new devices or infrastructure.
  2. Correlate reads with egress and rotate affected credentials after confirming exposure.
03Inference access is proxiedBedrock discovery / key testingNative logs · Draft

Authorized model access is resold through attacker reverse proxies.

What to look for

A key enumerates Bedrock models from an unfamiliar origin, followed by unexpected model access or usage.

Visibility: Defender-side logs where collected

Required logs

  • AWS CloudTrail management events
  • Bedrock invocation logging / relevant data events when explicitly enabled
  • Key inventory and inference usage

Coverage limit. The starter queries cover discovery only. Do not assume default CloudTrail management logs contain InvokeModel / Converse requests or prompt content.

AWS discovery candidates: Sigma CloudTrail, Sentinel AWSCloudTrail, and Splunk aws:cloudtrail. Invocation logging is a separate prerequisite.

Sigma
Draft SigmaDownload Sigma
title: Hunt - Bedrock discovery / key testing
id: 89e0e836-0c7b-44b4-a9d4-3524f68318b2
status: experimental
description: A key enumerates Bedrock models from an unfamiliar origin, followed
  by unexpected model access or usage.
author: MagicSword OpenAttacks catalog
date: 2026-10-23
falsepositives:
  - Authorized activity matching this broad hunt; baseline and investigate
    before alerting.
level: low
logsource:
  product: aws
  service: cloudtrail
detection:
  selection:
    eventSource: bedrock.amazonaws.com
    eventName:
      - ListFoundationModels
      - GetFoundationModel
  condition: selection
KQL
Draft KQLDownload KQL
// Discovery only; correlate with principal inventory and enabled invocation logs.
AWSCloudTrail
| where TimeGenerated > ago(7d)
| where EventSource == "bedrock.amazonaws.com"
| where EventName in ("ListFoundationModels", "GetFoundationModel")
| summarize Requests=count() by UserIdentityArn, SourceIpAddress, UserAgent, EventName, bin(TimeGenerated, 1h)
| order by Requests desc
SPL
Draft SPLDownload SPL
sourcetype="aws:cloudtrail" eventSource="bedrock.amazonaws.com"
eventName IN ("ListFoundationModels", "GetFoundationModel")
| bin _time span=1h
| stats count by _time userIdentity.arn sourceIPAddress userAgent eventName
| sort - count
Investigation steps
  1. Group ListFoundationModels / GetFoundationModel calls by key, source IP and user agent; compare against normal application identities.
  2. Correlate with invocation logs and cost. Discovery alone does not establish a stolen key or successful inference.
04Victim accounts incur inference usageUnexpected cloud / inference identity useCustom schema · Draft

The report describes high-volume requests and potentially substantial costs.

What to look for

A service key or role is used from an unfamiliar origin, gains broader access or causes inference usage outside its established baseline.

Visibility: Defender-side logs where collected

Required logs

  • Cloud control-plane and identity audit
  • AI-service invocation / diagnostic logs when enabled
  • Usage, quotas, cost and key inventory

Coverage limit. Cloud control-plane logs alone may not include model prompts or inference calls. No endpoint policy is claimed to stop cloud-only key misuse.

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 cloud / inference identity use
id: afac2e65-c838-485b-a26f-7f175d422f61
status: experimental
description: A service key or role is used from an unfamiliar origin, gains
  broader access or causes inference usage outside its established baseline.
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:
      - cloud.credential.use
      - cloud.role.change
      - model.invoke
  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 ("cloud.credential.use", "cloud.role.change", "model.invoke")
| project TimeGenerated, EventType, Actor, Asset, Action, Target, CorrelationId
| order by TimeGenerated desc
SPL
Draft SPLDownload SPL
sourcetype="openattacks:audit"
EventType IN ("cloud.credential.use", "cloud.role.change", "model.invoke")
| table _time EventType Actor Asset Action Target CorrelationId
| sort - _time
Investigation steps
  1. Pivot on key ID / principal, source network and user agent, then inspect role changes, token minting and service pivots.
  2. Join inference usage and billing anomalies to the same principal; verify whether the behavior was authorized.
Technical detailsTooling, impact & MITRE ATLAS

Attack context

AI's role
AI credential theft
Initial access
Vulnerable Laravel (CVE-2021-3129) leading to cloud credential theft

Tooling as reported

AI products, models and artifacts

Other tooling

  • LLM reverse proxy for resale

Targets and reported impact

Reported scale / exposure
Potential >$46,000/day Claude 2.x costs cited; July 2024 spike 61k Bedrock requests in 3 hours

Detection rules

  • Monitor AWS Bedrock API abuseGUIDANCE

    Alert on unusual Bedrock ListFoundationModels, InvokeModel, and Converse API calls, especially from AKIA long-term keys. Monitor for spike patterns (e.g., 61k requests in 3 hours). Implement CloudTrail analysis for Bedrock usage from unexpected geolocations or IAM principals.

  • Detect stolen cloud credentialsGUIDANCE

    Scan for exposed AWS keys in public repositories, logs, and error tracking services. Rotate credentials immediately upon detection. Use SCPs to restrict Bedrock access to approved principals only.

Activity timeline

First seen
2024-05-06
Last updated
2026-11-05

Updates

  • 2024-05-06 — Sysdig

    Initial LLMjacking disclosure; Laravel CVE-2021-3129 to cloud credential theft and AI inference resale

  • 2024-07-01 — Sysdig

    Bedrock Converse API abuse observed; 85k+ requests within 30 days of launch

  • 2026-01-28 — Pillar Security

    Bizarre Bazaar campaign targets AI startups with LLMjacking

  • 2026-09-03 — FortiGuard

    AWS Bedrock Marketplace abuse; attackers subscribe victims to high-cost models without approval

  • 2026-09-09 — Okta Security

    Stolen AI session tokens enable LLMjacking without long-term credentials; bypasses MFA and key rotation

Source defensive notes

Monitor Bedrock ListFoundationModels+Invoke/Converse from long-term AKIA keys; rotate exposed keys; SCP restrict Bedrock

MITRE ATLAS

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

AML.T0012Valid Accounts

Stolen cloud credentials are used to access inference services. Source

Sources & research credit

3 sources

Research by Alessandro Brucato / Sysdig TRT · Sysdig

  1. Sysdig: LLMjacking - Stolen Cloud Credentials Used in New AI Attack www.sysdig.com
  2. Sysdig: Growing Dangers of LLMjacking www.sysdig.com
  3. Pillar Security: Bizarre Bazaar - LLMjacking Campaign Targeting AI Startups www.pillar.io

2026 updates expand LLMjacking scope: Bedrock Marketplace abuse (FortiGuard), AI startup targeting (Pillar Bizarre Bazaar), and session token theft (Okta)

Record & downloads
Source record name
LLMjacking via stolen cloud credentials (Sysdig)
Entry ID
llmjacking-sysdig-2024-05
Catalog review
November 5, 2026
Category
attacks-on-ai
AI involvement confidence
confirmed