Skip to navigation

Threat reportCloud SecurityTL-2026-6293

Cloud Keys Leaked in Docker Hub Images Stolen Within Seconds by Automated 'ContainerBots' Scanners

highACTIVE

Stony Brook University researchers published 300 honeypot images across eight public container registries embedding AWS, OpenAI and Anthropic keys plus decoy SSH/database credentials. Planted AWS keys were first used 7.86 seconds after publication (median 51.19 s) and all 100 AWS keys were abused within 15 minutes, indicating automated static-analysis harvesting of image layers and registry APIs.

Key facts

IdentifiersTL-2026-6293
Also known asContainerBots
SeverityHIGH
First disclosed
Affected productsDocker Docker Hub; Harbor (CNCF) Harbor container registry; Amazon Web Services AWS access keys embedded in container images; OpenAI / Anthropic API keys embedded in container images
Weaknesses (CWE)CWE-798, CWE-312
ToolsContainerBots
MITRE ATT&CK techniquesT1078.004 Cloud Accounts, T1119 Automated Collection, T1496.004 Cloud Service Hijacking, T1526 Cloud Service Discovery, T1552.001 Credentials In Files, T1580 Cloud Infrastructure Discovery, T1589.001 Credentials, T1593.003 Code Repositories
MotivationUNKNOWN
Target sectorstechnology, cloud services, software development, finance
Target regionsGlobal
Detection rules9 (1 shown free)
Sources6 cited sources

Detection coverage

As of 2026-11-18, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-6293 across Splunk SPL, Microsoft KQL and Sigma, covering 10 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: ContainerBots Leaked AWS Key Validation Then Enumeration From External IP (TL-2026-6293)

Detects the observed ContainerBots post-theft progression on a long-term AWS access key: sts:GetCallerIdentity validity check followed by List/Describe enumeration from a non-corporate, non-AWS source IP within 15 minutes. Research showed all 100 planted AWS keys were abused within 15 minutes of image publication (first use 7.86 s, median 51.19 s), mostly key validation and resource listing. Maps to Valid Accounts: Cloud Accounts (T1078.004), Cloud Service Discovery (T1526) and Cloud Infrastructure Discovery (T1580). Public SPL detection for this threat.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1078.004, T1526, T1580
  • Data sources: aws:cloudtrail (index=aws), AWSCloudTrail (Microsoft Sentinel), AWS CloudTrail logsource (Sigma)
  • Severity: high
  • Confidence: high

Splunk SPL

index=aws sourcetype="aws:cloudtrail" earliest=-4h userIdentity.type=IAMUser userIdentity.accessKeyId=AKIA*
| eval access_key='userIdentity.accessKeyId', iam_user='userIdentity.userName'
| where NOT (cidrmatch("10.0.0.0/8",sourceIPAddress) OR cidrmatch("172.16.0.0/12",sourceIPAddress) OR cidrmatch("192.168.0.0/16",sourceIPAddress) OR like(sourceIPAddress,"%.amazonaws.com") OR sourceIPAddress=="AWS Internal")
| eval is_validate=if(eventName=="GetCallerIdentity",1,0)
| eval is_enum=if(match(eventName,"^(List|Describe|GetAccountAuthorizationDetails|GetAccountSummary)"),1,0)
| stats sum(is_validate) AS validate_calls sum(is_enum) AS enum_calls dc(eventSource) AS services dc(eventName) AS distinct_apis values(eventName) AS apis min(_time) AS first_seen max(_time) AS last_seen BY access_key iam_user sourceIPAddress
| eval window_secs=last_seen-first_seen
| where validate_calls>=1 AND enum_calls>=5 AND window_secs<=900
| convert ctime(first_seen) ctime(last_seen)
| table first_seen last_seen access_key iam_user sourceIPAddress validate_calls enum_calls services distinct_apis apis

Microsoft KQL

AWSCloudTrail
| where TimeGenerated > ago(4h)
| where UserIdentityType == "IAMUser" and UserIdentityAccessKeyId startswith "AKIA"
| where ipv4_is_private(SourceIpAddress) == false and SourceIpAddress !endswith ".amazonaws.com" and SourceIpAddress != "AWS Internal"
| extend IsValidate = EventName == "GetCallerIdentity"
| extend IsEnum = EventName matches regex "^(List|Describe|GetAccountAuthorizationDetails|GetAccountSummary)"
| summarize FirstSeen = min(TimeGenerated), LastSeen = max(TimeGenerated), ValidateCalls = countif(IsValidate), EnumCalls = countif(IsEnum), Services = dcount(EventSource), Apis = make_set(EventName, 50) by UserIdentityAccessKeyId, UserIdentityUserName, SourceIpAddress
| where ValidateCalls >= 1 and EnumCalls >= 5 and (LastSeen - FirstSeen) <= 15m
| project FirstSeen, LastSeen, UserIdentityAccessKeyId, UserIdentityUserName, SourceIpAddress, ValidateCalls, EnumCalls, Services, Apis

Sigma

title: ContainerBots Leaked AWS Key - GetCallerIdentity By Long-Term IAM Key From External IP
id: 3f6a1c0e-6293-4a11-9b7d-0a1e5c2d7001
name: tl_6293_aws_key_validate
status: experimental
description: Long-term IAM access key (AKIA) validated with sts:GetCallerIdentity from a non-corporate, non-AWS source IP. First step of the ContainerBots-style post-theft progression (key validity check) in TL-2026-6293.
references:
  - https://www.bleepingcomputer.com/news/security/cloud-keys-leaked-in-docker-hub-images-are-stolen-within-seconds/
  - https://containerbots.github.io
  - https://attack.mitre.org/techniques/T1078/004/
author: Threadlinqs Team
date: 2026-11-18
tags:
  - attack.initial-access
  - attack.t1078.004
logsource:
  product: aws
  service: cloudtrail
detection:
  selection:
    eventSource: 'sts.amazonaws.com'
    eventName: 'GetCallerIdentity'
    userIdentity.type: 'IAMUser'
    userIdentity.accessKeyId|startswith: 'AKIA'
  filter_internal:
    sourceIPAddress|cidr:
      - '10.0.0.0/8'
      - '172.16.0.0/12'
      - '192.168.0.0/16'
  filter_aws_service:
    sourceIPAddress|endswith: '.amazonaws.com'
  filter_aws_internal:
    sourceIPAddress: 'AWS Internal'
  condition: selection and not 1 of filter_*
falsepositives:
  - Developers validating their own keys from home or VPN egress ranges
  - CI/CD runners outside AWS using long-term keys
level: informational
---
title: ContainerBots Leaked AWS Key - Read-Only Resource Enumeration By Long-Term IAM Key From External IP
id: 3f6a1c0e-6293-4a11-9b7d-0a1e5c2d7002
name: tl_6293_aws_key_enumeration
status: experimental
description: List/Describe enumeration APIs called with a long-term IAM key from a non-corporate, non-AWS source IP. Building block for the TL-2026-6293 correlation.
references:
  - https://containerbots.github.io
  - https://attack.mitre.org/techniques/T1580/
author: Threadlinqs Team
date: 2026-11-18
tags:
  - attack.discovery
  - attack.t1580
  - attack.t1526
logsource:
  product: aws
  service: cloudtrail
detection:
  selection:
    userIdentity.type: 'IAMUser'
    userIdentity.accessKeyId|startswith: 'AKIA'
    eventName|startswith:
      - 'List'
      - 'Describe'
      - 'GetAccountAuthorizationDetails'
      - 'GetAccountSummary'
  filter_internal:
    sourceIPAddress|cidr:
      - '10.0.0.0/8'
      - '172.16.0.0/12'
      - '192.168.0.0/16'
  filter_aws_service:
    sourceIPAddress|endswith: '.amazonaws.com'
  filter_aws_internal:
    sourceIPAddress: 'AWS Internal'
  condition: selection and not 1 of filter_*
falsepositives:
  - Cloud inventory and CSPM tooling using static keys from external SaaS ranges
  - Administrators browsing via external networks
level: informational
---
title: ContainerBots Leaked AWS Key - Key Validation Followed By Enumeration Within 15 Minutes
id: 3f6a1c0e-6293-4a11-9b7d-0a1e5c2d7003
status: experimental
description: Same long-term IAM access key validated via GetCallerIdentity and then used for enumeration from outside the corporate/AWS network within 15 minutes. Mirrors the observed ContainerBots progression (check key, list resources, sometimes consume) from TL-2026-6293. Event-count thresholds are enforced in the SPL and KQL flavors.
references:
  - https://containerbots.github.io
author: Threadlinqs Team
date: 2026-11-18
tags:
  - attack.initial-access
  - attack.discovery
  - attack.t1078.004
  - attack.t1580
  - attack.t1526
correlation:
  type: temporal_ordered
  rules:
    - tl_6293_aws_key_validate
    - tl_6293_aws_key_enumeration
  group-by:
    - userIdentity.accessKeyId
  timespan: 15m
falsepositives:
  - Newly onboarded automation that validates credentials before inventory runs
  - Penetration tests and red-team exercises using issued keys
level: high

Known false positives:

  • Developers or CSPM/inventory tools using static IAM keys from external or VPN egress IPs
  • CI/CD runners outside AWS that validate credentials and then list resources

Tuning notes: Replace the RFC1918 exclusions with your corporate and VPN egress CIDRs (a lookup of known egress IPs is better). Exclude known automation access keys. Raise enum_calls or shorten the window to cut noise; lower them for keys that should never be used from outside the VPC. Any hit on a key that was ever embedded in a container image should be treated as compromise: revoke and rotate the key.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • ContainerBots Established AWS Key Used From Never-Before-Seen External IP (TL-2026-6293) · unlock with TL Creds
  • ContainerBots Compute Resource Consumption With Long-Term AWS Key From External IP (TL-2026-6293) · unlock with TL Creds
  • ContainerBots Foreign Pull Within Seconds Of Image Push (TL-2026-6293) · unlock with TL Creds
  • ContainerBots Exposure: Container Build/Run/Push With Embedded Cloud or AI Credentials (TL-2026-6293) · unlock with TL Creds
  • ContainerBots LLM Service Hijacking: Bedrock Access With Long-Term AWS Key From External IP (TL-2026-6293) · unlock with TL Creds
  • ContainerBots-Style Static Secret Extraction From Container Image Layers (TL-2026-6293) · unlock with TL Creds
  • ContainerBots Multi-Region AWS Resource Enumeration With Long-Term Key (TL-2026-6293) · unlock with TL Creds
  • ContainerBots Registry API Enumeration For Recently Published Images (TL-2026-6293) · unlock with TL Creds

Indicators of compromise

As of 2026-11-18, Threadlinqs Intelligence tracks 10 indicators of compromise for TL-2026-6293.

Sample indicators (3 of 10)

  • Domain: tsouvalas[.]xyz
  • URL: hxxps://www[.]bleepingcomputer[.]com/news/security/cloud-keys-leaked-in-docker-hub-images-are-stolen-within-seconds/
  • Domain: securitee[.]org

Defanged, TLP:CLEAR. Full set of 10 indicators: unlock with TL Creds.

10 indicators of compromise · unlock with TL Creds. Compare plans

Technical analysis

Researchers Billy Tsouvalas and Nick Nikiforakis (Stony Brook University) presented a four-month honeypot study at the ACM Conference on Computer and Communications Security (CCS) on 2026-11-16, reported by BleepingComputer on 2026-11-17. They published 300 honeypot container images across eight public container registries (Docker Hub and Harbor are named in the coverage; the other registries are not listed in the sources reviewed). The images embedded planted AWS, OpenAI and Anthropic API keys plus decoy SSH and database credentials, and the researchers deployed detection that would alert when an image was built, run, or had a shell opened.

The images were pulled more than 75,000 times. The planted AWS keys received over 140,000 API calls from about 5,000 unique IP addresses. All 100 planted AWS keys were used within 15 minutes of publication; the fastest first use came 7.86 seconds after publication and the median time to first API call was 51.19 seconds. Observed AWS activity ranged from checking whether the keys worked, to listing available resources, to sometimes consuming them. OpenAI and Anthropic keys were quickly used to exhaust test-account balances. Docker Hub accounted for nearly all AWS key activity; Harbor also saw some AWS key abuse.

The scanners were collectively labelled 'ContainerBots', a loose umbrella term: the researchers found no evidence that the scanning was coordinated by the same operators. Because the monitoring rarely detected anyone running the images, the researchers concluded the scanners overwhelmingly extract credentials statically, by unpacking image layers or querying registry APIs, without starting containers. No CVE, malware family, IP list or other indicator list is given in the coverage reviewed, so no network IOCs were available to correlate against BeaconBeagle.

Disclosure: the findings and full paper were reported to Docker in May 2026; Docker sent an acknowledgment and nothing further. A small-scale follow-up test in August 2026 showed no changes in the platform's defenses. The researchers recommend that Docker Hub scan images for secrets before they go live and warn publishers, and that access to the public feed of recently published or updated images be restricted or rate-limited to slow mass-scanning. Earlier independent work shows the exposure is widespread: an RWTH Aachen study (ASIA CCS 2023) found about 8.5% of 337,171 Docker Hub images and 8,076 private-registry images contained secrets, and Flare (December 2025) found 10,456 Docker Hub images uploaded in one month (November 2025) containing exposed keys, with roughly 75% of affected developers not revoking the underlying credential.

MITRE ATT&CK techniques by tactic:

Initial Access

T1078.004 Cloud Accounts

Collection

T1119 Automated Collection

Impact

T1496.004 Cloud Service Hijacking

Discovery

T1526 Cloud Service Discovery; T1580 Cloud Infrastructure Discovery

Credential Access

T1552.001 Credentials In Files

Reconnaissance

T1589.001 Credentials; T1593.003 Code Repositories

Remediation

Immediate actions

  • Treat any credential ever embedded in a pushed container image as compromised: revoke and rotate it immediately, not just delete it from the image or tag
  • Audit public and shadow-IT (personal/contractor) registry accounts for images containing AWS, OpenAI, Anthropic, SSH or database credentials
  • Review AWS CloudTrail, and OpenAI/Anthropic usage and billing, for unexpected calls or spend from unfamiliar IPs shortly after any image publication

Workarounds

  • Use private repositories for images that must contain configuration
  • Set spend limits and usage alerts on AI API accounts and scope keys to least privilege

Longer-term hardening

  • Keep secrets out of images and manifests; inject them at runtime from a secrets manager (e.g. AWS Secrets Manager, HashiCorp Vault, Azure Key Vault)
  • Replace long-lived static keys with short-lived credentials (SSO, STS AssumeRole, workload identity)
  • Run secret scanning in CI before images are pushed and continuously across registries, repos and build logs
  • Registry operators: scan images for secrets before they go live, warn publishers, and restrict or rate-limit the public feed of recently published or updated images

Affected and fixed versions

  • Docker — Docker Hub
    Vulnerable versions: Public registry; no pre-publication secret scanning observed in the August 2026 follow-up test
  • Harbor (CNCF) — Harbor container registry
    Vulnerable versions: Public instances; some AWS key abuse observed
  • Amazon Web Services — AWS access keys embedded in container images
    Vulnerable versions: Any key published in a public image
  • OpenAI / Anthropic — API keys embedded in container images
    Vulnerable versions: Any key published in a public image

Timeline

  • Trend Micro publishes honeypot research showing DockerHub credentials leaking from TeamTNT-controlled accounts (separate, earlier registry-credential exposure; context only).
  • RWTH Aachen presents 'Secrets Revealed in Container Images' at ASIA CCS '23: about 8.5% of 337,171 Docker Hub images and 8,076 private-registry images contained secrets.
  • Flare reports 10,456 Docker Hub images uploaded in November 2025 contained exposed keys; about 75% of affected developers did not revoke the credential.
  • Stony Brook researchers report findings and the full paper to Docker (exact day in May not stated); Docker sends an acknowledgment and no further reply. The 4-month, 300-image honeypot experiment preceded this report.
  • Small-scale follow-up test (exact day in August not stated) shows no changes to Docker Hub defenses against secret harvesting.
  • Findings presented at ACM CCS: 300 honeypot images across eight registries pulled 75,000+ times; first AWS key use 7.86 s after publication, median 51.19 s, all 100 AWS keys used within 15 minutes.
  • BleepingComputer publishes coverage of the research; 140,000+ AWS API calls from about 5,000 unique IPs and OpenAI/Anthropic balance exhaustion reported.

Sources cited

Drafted by the Threadlinqs research pipeline from 6 sources; published ; last updated —.

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats