A cybersecurity incident involving Belfast-based technology provider Wellington IT has exposed sensitive financial information associated ...
Attackers are extracting and exploiting exposed cloud credentials from publicly accessible Docker images within seconds of their publication, according to a four-month experiment conducted by researchers at Stony Brook University. The fastest observed exploitation occurred just 7.86 seconds after an image containing a planted AWS key was uploaded to Docker Hub.
Researchers Billy Tsouvalas and Nick Nikiforakis investigated the activity by publishing 300 instrumented honeypot images across eight public container registries. The experiment examined how quickly automated scanners discover credentials embedded in container images and whether they subsequently attempt to use them.
Conducted over four months, the study involved planting AWS, OpenAI, and Anthropic API keys alongside decoy SSH credentials and database passwords. The researchers also deployed monitoring mechanisms designed to detect different interactions with their images, including image builds, container execution, and shell access.
The findings, presented at the 2026 ACM SIGSAC Conference on Computer and Communications Security (CCS), revealed extensive automated credential harvesting.
AWS Credentials Exploited Within Seconds
Across the experiment, the researchers recorded more than 75,000 image pulls. The planted AWS credentials alone attracted over 140,000 API requests originating from approximately 5,000 unique IP addresses.
The observed requests ranged from preliminary authentication checks to attempts to enumerate available cloud resources. Some interactions progressed further, indicating attempts to consume resources using the exposed credentials.
The earliest recorded AWS API request occurred just 7.86 seconds after an image was published on Docker Hub. The median time between publication and the first API request was 51.19 seconds. Furthermore, all 100 planted AWS keys had been accessed within 15 minutes of publication.
The OpenAI and Anthropic credentials experienced similar abuse. Automated activity rapidly consumed the balances associated with the researchers' test accounts, demonstrating that exposed AI service tokens can also become targets for immediate financial exploitation.
Docker Hub generated almost all observed AWS credential activity, although the researchers also recorded instances of abuse involving Harbor, another container registry included in the experiment.
Automated Scanners Extract Credentials Without Running Containers
The researchers collectively referred to the automated scanning activity as "ContainerBots." However, the designation does not establish that all observed scanners belonged to one threat actor or operated through a centrally coordinated infrastructure.
An important finding concerned how these bots accessed the exposed credentials.
Monitoring systems rarely detected attempts to execute the planted containers, suggesting that attackers generally did not need to run an image to retrieve its contents. Instead, they relied on static inspection techniques, including unpacking image layers and querying registry APIs to retrieve filesystem contents.
This approach allows scanners to examine files stored within container images without launching the applications themselves.
Container images package application code, dependencies, configuration files, and other resources into reusable layers. Consequently, credentials accidentally included during image creation can remain accessible through those layers even when developers do not intend to distribute them.
The researchers explained that their observations overwhelmingly involved static analysis rather than actual container execution.
Docker Hub Secret Exposure Is an Established Problem
The latest findings add to previous research documenting widespread credential exposure across public container repositories.
In December 2025, researchers at cybersecurity company Flare identified 10,456 Docker Hub images containing exposed secrets. These included approximately 4,000 AI-related API keys, alongside credentials capable of providing access to cloud environments, databases, Git repositories, and CI/CD infrastructure.
An earlier 2023 investigation involving 337,171 Docker images identified 52,107 valid private keys and 3,158 distinct API secrets across 28,621 images.
Such exposures create substantial remediation challenges. Removing an affected image or deleting a credential from its configuration does not invalidate copies already retrieved by automated scanners. Organizations must revoke compromised credentials, issue replacements, examine access logs, and assess whether unauthorized access occurred.
Researchers Recommend Additional Docker Hub Defenses
Tsouvalas and Nikiforakis recommend introducing automated secret detection before container images become publicly accessible. Docker Hub could scan uploaded images for embedded credentials and notify publishers when potentially sensitive information is discovered.
They also propose restricting or rate-limiting access to Docker Hub's public feed of recently published or updated images, making large-scale automated collection more difficult.
The researchers told BleepingComputer that they submitted their findings and the complete research paper to Docker in May 2026. Although the company acknowledged receiving the disclosure, they reportedly received no subsequent response.
A smaller experiment conducted in August to determine whether Docker had introduced additional protections did not identify any changes.
Docker's existing documentation already recommends using BuildKit secret mounts rather than passing credentials through Dockerfile "ARG" or "ENV" instructions, which can leave sensitive information in image contents. Secret mounts make credentials temporarily available during specific build operations without automatically persisting them in the resulting image.
The National Institute of Standards and Technology (NIST), in its SP 800-190 Application Container Security Guide, similarly recommends keeping secrets outside container images and supplying them dynamically when required.
The research demonstrates that publishing an image containing an active credential can expose it to automated collection almost immediately. Developers and organizations should therefore treat any credential embedded in a publicly accessible image as compromised, revoke it promptly, and ensure that replacement credentials are not retained in subsequent image layers.
Many smartphone users are experiencing unwanted calls from unknown numbers, ranging from promotional messages and automated marketing calls to fraudster scams designed to steal money and personal information.
The majority of calls originate from databases containing telephone numbers collected from online and offline sources, although some are generated by random dialing systems. There is a growing market for the collection, organization, and sharing of personal information among businesses and third parties as consumer information becomes increasingly valuable.
When users register on a website, complete online forms, install mobile applications, shop online, or enter promotional contests, you can reveal your contact information to several organizations. There is also the possibility that this information may circulate through third-party data-sharing arrangements or be made available following a data breach.
As one of the lesser-known players in this ecosystem, data brokers are able to collect information from multiple sources and create consumer profiles that can then be sold to advertisers, marketing firms, and other interested parties. It is therefore possible to reach businesses and individuals through a number shared for ordinary services or registrations without any direct connection to the original source, for example. Cybercriminals can use such information to conduct targeted spam campaigns, impersonation scams, and financial fraud.
How Phone Numbers Find Their Way Into Spam Lists
The majority of spam calls are not the result of stolen databases. Some are reached using automated dialing systems that generate and call random combinations, while others are acquired from publicly available records, such as business directories, property documents, and online listings.
The commercial distribution of consumer information remains an important source of unwanted phone numbers, however. A number of methods are commonly employed to collect phone numbers, including registration forms, online purchases, loyalty programs, surveys, promotional offers, and account sign-ups.
If a user provides their contact information, the company may share it with advertising partners or other third parties in accordance with the company's data sharing policy. There are instances where consent is buried in lengthy terms and conditions or presented in confusing opt-in settings, causing confusion about where the information will end up. As part of this process, data brokers play a key role. They combine information gathered from various sources including websites, public records, purchase histories, social media activity, and commercial databases to create consumer profiles.
Marketing agencies, businesses, and other organizations can purchase or license the resulting lists. Despite the fact that most of this activity occurs within the advertising industry, poorly controlled data-sharing practices can also make contact information available to scammers. It is also possible to expose customer information through breaches of security and phishing attacks.
When a security breach occurs, customer records may include telephone numbers, names, email addresses, and other personal information when an online service suffers a security incident. As an alternative to using legitimate websites, cybercriminals can also retrieve information through the use of fake job advertisements, fake login pages, and malicious applications.
By using automated tools known as web scrapers, some phone numbers can also be harvested directly from websites and social media platforms. Once a number has been scanned and extracted in bulk, it can be added to calling lists or combined with other personal information to make fraudulent calls appear more convincing.
Although it is important to distinguish between legitimate data collection and criminal activity, these two concepts can intersect when personal information is shared widely, inadequately protected, or exposed by security flaws.
Ultimately, a number that was originally provided to register for a shopping account or service may become part of a database that is used to sell unsolicited information or perpetrate targeted scams.
How to Keep a Phone Number Away From Spammers
A person can minimize their chances of receiving unwanted calls and targeted scams by limiting the exposure of their personal contact information, even though preventing every spam call may not be possible. The first step to preventing unwanted calls and targeted scams is to be selective when sharing personal contact information. It is important to be cautious when dealing with unfamiliar websites, suspicious registration pages, fraudulent job portals, and promotional offers requesting unnecessary personal information.
When submitting information to a website, you should check the reputation and privacy policies to identify questionable services. Public exposure should also be considered. It is important to remove phone numbers from public social media profiles, online directories, and other platforms whenever possible. Mobile users should review application permissions on a regular basis, particularly access to contact information, location, photographs, and other personal information.
A user can revoke unnecessary permissions through the privacy settings on Android and iOS devices. Reviewing account privacy options as well as avoiding unnecessary third party sign-ups can also help limit data sharing. For users already receiving frequent unwanted calls, built-in spam detection features and call-blocking services can assist in filtering out suspicious numbers.
In order to avoid answering unknown calls immediately, you can place them into voicemail. It is important to avoid giving passwords, banking information, OTPs, or remote access to devices to suspicious callers, regardless of their claim to represent a trustworthy organization. A user should also check whether their personal information has been exposed in known data breaches.
Services such as Have I Been Pwned can be used to identify exposed email addresses corresponding to compromised accounts, while searching online may reveal information that can be accessed by the public. The primary cause of spam calls is not always careless sharing. Random dialing, commercial data collection, and security breaches can expose numbers without the owner being aware of them.
Spammers and cybercriminals may have a harder time converting ordinary contact information into opportunities for fraud if privacy practices are strengthened, app permissions are handled carefully, and unsolicited calls are made more aware.