In the modern digital economy, the cloud is the backbone of infrastructure. Yet, beneath the polished interface of major cloud service providers lies a persistent, festering security crisis: the proliferation of forgotten, exposed, and highly privileged AWS access keys. New research from Truffle Security has revealed a staggering reality—tens of thousands of corporate AWS access keys, many of which carry full administrative permissions, remain active and valid, years after they were first exposed to the public internet.
This discovery highlights a critical failure in lifecycle management within corporate IT departments. Despite repeated warnings from cybersecurity experts, the "set it and forget it" mentality surrounding cloud credentials has created an environment where attackers do not need to hunt for vulnerabilities—they simply need to search public repositories.
The Scale of the Exposure: A Statistical Audit
Truffle Security’s comprehensive analysis, which spanned data surfaced between August 2022 and August 2026, identified 64,024 unique AWS key pairs across 431,875 public findings. The breadth of these leaks is alarming, spanning across git history, Docker images, public package registries, and, most notably, datasets hosted on platforms like Hugging Face.
When researchers narrowed their scope to 10,616 keys that featured complete, valid credentials, the results were sobering: 88% of these keys remained active and fully functional. Even more concerning was the discovery of 526 root keys—the "master keys" to an entire AWS account—and 242 IAM (Identity and Access Management) users holding AdministratorAccess policies. These users possess the power to modify, delete, or exfiltrate any data within the compromised environment, effectively granting an attacker total control.
The Hugging Face Factor
The research specifically highlighted Hugging Face as a primary vector for these leaks. With 8,482 unique live keys found across 3,394 public datasets, it has become a hotspot for accidental credential exposure. Researchers noted that the nature of these datasets—often snapshots of public code repurposed for AI model training—makes them a particularly dangerous repository for secrets. Approximately 17.9% of the keys discovered on Hugging Face were root keys, representing the highest density of privileged exposure among all sources tracked by the firm.
A Chronology of Neglect: The "Zombie" Key Phenomenon
Perhaps the most damning aspect of the report is the age of the exposed credentials. The median age of these live, active keys is 1,831 days—roughly five years. The oldest key identified in the study dates back 17.4 years, nearly as old as the AWS IAM service itself.
The data suggests that these are not recent accidents but long-standing, unaddressed security failures. Only 25 of the thousands of keys analyzed (0.9%) were generated within the last 30 days. This indicates that the vast majority of these credentials have been sitting exposed in public view for years, effectively acting as "zombie" accounts that organizations have forgotten to rotate or deactivate.
The Failure of Rotation
Standard security practice dictates that access keys should be rotated regularly—a process of replacing an old key with a new one and retiring the former. Truffle Security found that 86% of the exposed keys they analyzed had never been rotated or superseded. In instances where the firm could identify the associated user, only 13.7% possessed a newer, legitimate key alongside the leaked one. This indicates a systemic failure to implement basic credential hygiene, leaving the original, compromised keys to remain valid and available for malicious actors.
Supporting Data: Quarantine and Oversight
The research also shed light on the efficacy of AWS’s own mitigation efforts. Of the 7,590 active IAM users identified, 929 were already carrying the AWSCompromisedKeyQuarantine policy—a mechanism applied by AWS when it detects that credentials have been exposed.
However, the findings reveal a disturbing gap in enforcement. While these users were flagged by AWS and supposedly restricted, they remain active. Even more concerning, 112 of these users were operating under an "original version" of the quarantine policy that AWS officially stopped applying in 2023. These users were identified and notified at least three years ago, yet no corrective action was taken by the organizations responsible. The keys remained authenticated, and the "quarantine" was effectively treated as a suggestion rather than a mandate.
The Implications for Global Security
The implications of this data extend far beyond simple data theft. The presence of these keys represents a massive, unmitigated attack surface for:
- Cryptojacking: With 90.5% of the accounts associated with these leaked keys lacking any budget alarms, attackers can spin up massive compute resources to mine cryptocurrency, leaving the account holder with a staggering, unexpected bill.
- Data Exfiltration: With full
AdministratorAccess, an attacker can bypass all perimeter defenses to siphon sensitive intellectual property, proprietary datasets, or customer PII (Personally Identifiable Information). - Supply Chain Contamination: Because these keys are often found in CI/CD logs and Docker images, they can be used to inject malicious code into the software supply chain, affecting not just the company that leaked the key, but every downstream user of their software.
- AI Model Poisoning: Given the high concentration of keys on platforms like Hugging Face, there is a clear risk of malicious actors gaining access to the infrastructure used to train or host AI models, potentially introducing backdoors or bias into the models themselves.
Official Responses and Industry Guidance
In response to the findings, an AWS spokesperson emphasized the company’s proactive measures, stating: "Any time AWS is aware of exposed keys, we notify the affected customers. We also thoroughly investigate all reports of exposed keys and quickly take any necessary actions, such as applying quarantine policies to minimize risks for customers without disrupting their IT environment." The statement further encouraged customers to reach out to AWS Support for guidance on securing their accounts.
However, security researchers argue that the responsibility lies ultimately with the organizations. Truffle Security suggests that the existence of root access keys in 2026 is entirely unjustifiable and that these should be deleted immediately. Furthermore, they advocate for a cultural shift: security teams must operate under the assumption that if a secret has been committed to a repository, it is already compromised. Deleting a file from a repository history is often ineffective once the data has been ingested into training corpora or mirrored in other locations.
Moving Toward Tighter Access Controls
To combat this, the report outlines several critical steps for IT and security leaders:
- Immediate Purge of Root Keys: Organizations should audit their IAM users and ensure that root credentials are not being used for daily operational tasks.
- Budget Alarms: Implementing even modest budget alerts can provide an early warning system for unauthorized compute usage, which is often the first sign of a compromised account.
- Automated Secret Scanning: Integrating robust secret-scanning tools into the CI/CD pipeline is essential. Waiting for a third party to discover a leak is a reactive strategy that is often too little, too late.
- Enforced Rotation Policies: Moving away from static, long-lived credentials toward short-lived, temporary tokens (such as those provided by AWS STS) reduces the "blast radius" of a potential leak.
- Proactive Lifecycle Management: Organizations must move beyond simply generating keys and start tracking them. Knowing which keys are in use, by whom, and for what purpose is the only way to ensure that deprecated credentials do not linger in the digital ether.
The findings from Truffle Security serve as a stark reminder that in the cloud era, security is not a one-time configuration but an ongoing lifecycle. As long as organizations continue to leave the keys to their digital kingdoms lying in public view, they remain vulnerable to the simplest of exploits. In an age where AI-driven cybercrime is accelerating, the failure to address these fundamental security hygiene issues is no longer just a technical oversight—it is a significant business risk.
About the Author
