Truffle Security’s Secrets Revocation Program and exposed credentials
A leaked API key is not fixed when someone sends an email about it. It is fixed when it stops working. That distinction sits at the center of Truffle Security’s new Secrets Revocation Program, a free initiative intended to help issuers invalidate keys found in public repositories, packages, container images, datasets and applications. Truffle Security launch post
Truffle Security says it has confirmed 1.4 million live secrets across public sources. That is the program’s target inventory—not a claim that all 1.4 million keys have already been revoked. The company’s launch post says it tested keys before delivering them to issuers, so the handoff focuses on credentials it found working at the time of verification. Truffle Security launch post
The stakes are not abstract. Truffle Security told BleepingComputer that 81% of the keys it reported the prior year were still working six months later. A notification does not disable a credential: it can be missed, ignored or left for an internal owner to handle. Until the issuer revokes the key, it remains usable. BleepingComputer report
An exposed credential can function as a legitimate key rather than trigger an obvious exploit alert. Finding a credential does not itself stop someone from using it. This article focuses on Truffle Security’s program and the process of invalidating exposed credentials, not on broader claims about AI systems.
Why detection has not been enough
A credential can surface in a public repository, a published package, a Docker image or an uploaded configuration file. Removing the original file does not necessarily disable copies already collected elsewhere, and finding the string does not tell us whether it still works. The response has to answer that second question—and, if the credential is live, get it revoked. Truffle Security launch post
Truffle Security reported finding 543,000 valid credentials in public GitHub repositories earlier in 2026 and more than 431,000 AWS secrets in a separate investigation spanning code repositories, Git histories, datasets, Docker images, registries and CI logs. These are distinct reported findings, not numbers to add together as a verified count of the program’s 1.4 million target keys. Nor do they show that those target keys have been revoked. They illustrate the problem the program is trying to address: detection can establish that a key is exposed and working, but a report alone leaves it active. BleepingComputer report Truffle Security launch post
Truffle Security also reported finding 2,655 working keys in configuration files for AI coding tools since January 2025, a type of finding the company said had not existed the previous year. BleepingComputer noted an incident in which exposed Hugging Face tokens were used by AI agents to steal platform data. The issue here is still the credential: once exposed, it needs to be disabled, not merely identified. BleepingComputer report
That is the gap Truffle’s program targets. A researcher can verify a key and report it, but the issuer controls the system that can invalidate it. Routing verified findings to that issuer connects discovery to the action that makes the credential stop working. Truffle Security launch post
How the revocation workflow works
Truffle Security says its TruffleHog tool scans public code, packages, container images, AI datasets and apps, among other sources, then tests discovered keys. According to the company, only keys that work at the time of testing are routed to the issuer. Its launch post reports that the research team scanned six petabytes of public data in October and more than 20 petabytes during the year. These are company-reported scanning figures, not a measure of unique credentials or proof that every public source has been searched. Truffle Security launch post
After verification, an issuer can participate in one of two ways described in the reporting. It can provide a revocation endpoint so the program can support automated invalidation, or receive verified findings through a webhook and revoke keys itself. BleepingComputer reports that the webhook route comes with an expectation that issuers revoke the disclosed keys within 14 days. That deadline is operationally important: a finding that lands in another inbox without an owner and response process risks recreating the same notification problem. BleepingComputer report Truffle Security launch post
This is not a promise that every key can be disabled without consequences. A credential may be tied to a production integration or a customer workflow, so issuers need to handle invalidation carefully and help affected users replace it. The available launch information describes the program’s revocation options, but does not establish a universal rotation procedure for every issuer or application. Teams should follow the issuer’s own guidance and test replacement credentials before depending on them. Truffle Security launch post
The distinction between disclosure and revocation also clarifies what the program does not claim. It does not make exposed code disappear from the internet, and it does not prove that no one used a key before it was disabled. Revocation cuts off future authentication with that credential; incident review may still be needed to determine whether access occurred and what data or services were affected. Truffle Security launch post
The issuer-led approach has precedent
BleepingComputer points to Google’s handling of exposed Google Cloud service-account keys as an example of what changes when the issuer acts. Truffle Security reported that when Google relied on notifying key owners, 72% of exposed secrets remained active. After Google began automatically disabling exposed keys by default, the figure fell to 11%. BleepingComputer reports that customers can opt out, so automatic disabling does not eliminate every valid exposed key. BleepingComputer report
The comparison gets at the difference between finding a key and making it unusable: a notification leaves the next step to the owner, while issuer-led revocation can disable the credential directly. The reported figures are specific to Google’s approach; they do not establish that every issuer will see the same result or that automatic revocation carries no risk. Issuers still have to account for service disruption and give customers a route to restore access with new credentials. BleepingComputer report
For organizations that issue API keys, this is a lifecycle problem as much as a scanning problem. Keys need controls at creation, storage, use, detection and retirement. A mature process should make it clear who receives verified reports, who can approve revocation, how customers are notified and how replacement credentials are issued. The details will vary by service, but leaving an active key in limbo is not a neutral choice.
What participation and the launch figures mean
Truffle Security said 16 issuers had joined at launch, naming Hugging Face, Notion, Vercel, Resend, RubyGems and Twilio among them. The company also reported that its pipeline held another 43,699 verified keys associated with issuers that had not yet joined. Those numbers describe the program’s position as reported at launch; they are not a current status dashboard, and they should not be read as proof that every listed key has since been revoked. Truffle Security launch post BleepingComputer report
Participation is free, according to Truffle Security. The company says issuers receive keys verified with TruffleHog before delivery and can contribute feedback on onboarding, delivery and revocation. That arrangement may reduce friction for issuers, but it still requires them to establish an accountable operational path for acting on findings. A free program cannot make an organization’s internal ownership decisions on its behalf. Truffle Security launch post
For security teams assessing the offer, useful questions follow naturally: Which key types are supported? What evidence accompanies each report? How is a mistaken or stale finding handled? Who receives webhook events, and how will the organization meet the 14-day expectation? What customer communication and replacement steps follow an invalidation? The source material does not answer every implementation detail, so issuers should settle those details during onboarding rather than assume the program handles them automatically. BleepingComputer report Truffle Security launch post
The central measure of this program is not how many secrets it finds, but whether issuers disable the verified keys they receive. Truffle’s reported counts describe different investigations and a larger target inventory; they do not establish that 1.4 million keys have been revoked. Its issuer-led model addresses the step between finding a live credential and shutting it off. That can close the notification gap, while leaving secret handling, least privilege, monitoring and incident response to the organizations that operate the systems. Truffle Security launch post BleepingComputer report
truffle securitys secrets revocation program and exposed credentials
truffle securitys secrets revocation program and exposed credentials
No-text editorial illustration of an exposed credential represented by a