# AI Cloud Vulnerabilities: Truffle Security’s Secrets Revocation Program Targets 1.4 Million Live Keys

- **Canonical URL:** https://www.probackend.com/api-key-security-lifecycle-risks/ai-cloud-vulnerabilities-truffle-security-s-secrets
- **Section:** Vulnerabilities & Exploits
- **Published:** 2026-11-17
- **Summary:** Truffle Security’s free program helps issuers verify and revoke exposed live credentials across cloud and AI infrastructure.

---

## Truffle Security targets 1.4 million live exposed secrets for revocation

Truffle Security’s free Secrets Revocation Program is intended to help credential issuers invalidate API keys and other secrets that have been exposed publicly but remain usable. The scale is notable: Truffle says it has identified 1.4 million publicly exposed secrets that are still valid. The program is a coordinated response to a gap that ordinary disclosure does not close—finding a credential is not the same as making it unusable.

The company says its discoveries span public code repositories, package releases, container images, datasets, and applications. These credentials can be exposed inadvertently during development: a developer may commit a configuration file, bake a token into a Docker image, or publish a package or dataset containing a secret. A credential can persist in Git history and copies of artifacts even after someone removes it from the current source tree. The immediate risk is not limited to the organization that issued the key; a compromised integration can affect customers, services, and downstream users that depend on it.

Truffle previously reported finding 543,000 valid credentials in public GitHub repositories, and more than 431,000 AWS secrets across code repositories, Git histories, datasets, Docker images, registries, and CI logs. Those figures describe separate research findings, not a claim that every exposed credential has been exploited. The central concern is that valid credentials remain an opportunity for misuse for as long as they work.

## What is a cloud vulnerability, and how do exposed secrets fit?

A cloud vulnerability is a weakness in cloud-hosted systems, services, configurations, software, or identity controls that could let an unauthorized party access data or perform actions. A leaked API key is not necessarily a software flaw in the cloud provider. It is an identity and access-control exposure: someone who obtains a still-valid secret may be able to act with the permissions attached to it. The severity therefore depends on the issuer, the key’s scope, the resources it can reach, and whether additional safeguards limit its use.

This distinction matters when discussing AI cloud vulnerabilities and broader cloud AI infrastructure vulnerabilities. AI applications often connect models and agents to cloud services through tokens, API keys, and service credentials. Exposure can turn those connections into an attack path, but the cited reporting does not establish that every leaked key belongs to an AI system or that each key has been abused. It does show why credential lifecycle controls are part of infrastructure security: secrets must be discovered, assessed, rotated or revoked, and removed from the places where they were copied.

Truffle Security reports that 81% of keys it notified organizations about the previous year still worked six months later. That statistic points to a practical weakness in notification-only remediation. Researchers can alert an owner, but they may not control the credential, and a busy or unresponsive owner can leave a known exposure active. This is an operational failure mode rather than proof that all notifications go unanswered.

## How the Secrets Revocation Program works

The program uses TruffleHog to check whether leaked keys are valid, then coordinates with participating issuers to revoke verified credentials. An issuer can provide a revocation endpoint for the program to use, or receive verified keys through a webhook and revoke them within 14 days, according to Truffle. This creates a path from detection to an action by the party able to disable the credential.

The issuer’s role is important. A researcher or scanner can identify a string that resembles a secret and, where supported, test whether it still works. Only the issuer can reliably determine the appropriate account and disable or replace the credential without unnecessarily disrupting legitimate use. Verification and issuer coordination can reduce false alarms and avoid treating every string found in a public artifact as an active compromise.

Sixteen issuers had joined when BleepingComputer reported the program’s launch. The named participants included Hugging Face, Notion, Vercel, Resend, RubyGems, and Twilio. Participation does not mean that every credential from those services is automatically revoked in all circumstances. The program’s stated aim is to enable issuer-side handling of verified exposed secrets, and outcomes depend on the issuer’s integration and applicable policies.

## Why automatic revocation can outperform notification alone

Google’s handling of exposed Google Cloud service-account keys provides a useful example cited in the report. Truffle said that when Google relied on notifying key owners, 72% of exposed secrets remained active. After Google began disabling exposed keys by default on June 16, 2024, Truffle found the active share had fallen to 11%. These numbers illustrate the potential effect of shifting from a request for customer action to an issuer-controlled default; they should not be read as a universal result for every provider or every type of secret.

Automatic revocation also involves tradeoffs. A key may support a production workload, and disabling it can cause an outage if the owner has not replaced it. Truffle acknowledges that customers can opt out, so the program cannot guarantee elimination of every valid exposed key. Issuers need processes for validating findings, communicating changes, managing exceptions, and helping customers restore service with a replacement credential. For consumers of the service, a revoked key is a security improvement only if legitimate systems can transition safely.

## AI development adds pressure to the secrets lifecycle

Truffle says it has found 2,655 working keys in configuration files used by AI coding tools since January 2025, a category it described as nonexistent the year before. This is evidence of a newer exposure pattern, not a measure of all AI-related credential leakage. Developers may connect coding assistants and related tools to repositories or services, creating additional configuration surfaces where credentials can appear.

The report also points to an incident in which AI agents used exposed Hugging Face tokens to steal data from the platform. This example illustrates how automation can increase the speed or scale at which an exposed credential is used. It does not establish that AI itself is the root cause of leaked secrets. The security issue remains that a usable credential was exposed and could be exercised; AI-enabled tools can change the context and potential impact of that exposure.

For teams building AI services, the practical implication is to treat secrets as managed identities, not static setup text. Keep credentials out of public repositories and distributed images, limit permissions, rotate them when exposure is suspected, and ensure that revocation does not depend solely on a developer seeing an alert. [Cloud and AI security programs](https://www.probackend.com/ai-corporate-data-breach/fireworks-ai-breach-widens-a-cloud-vendor-compromise-became) should also test how quickly a known exposed key can be disabled and how dependent workloads recover afterward.

## What issuers and customers can take from the program

API-key issuers can assess whether they can identify and revoke verified leaked credentials through a dedicated endpoint or another controlled integration. They should define verification thresholds, customer notification practices, service-impact safeguards, and an exception process. The program offers issuers a way to participate in coordinated revocation; it is not a substitute for an issuer’s own access controls or incident response.

Customers should assume that deleting a secret from a repository may not remove it from history, forks, package archives, container registries, logs, or cached copies. If exposure is suspected, revoke or rotate the credential, inspect its permissions and use, and replace it in dependent systems. Scanning helps find candidates, but a durable response requires both invalidation and cleanup, followed by controls that prevent the same class of accidental publication.

The wider lesson for cloud infrastructure security is that discovery without remediation leaves risk on the table. The reported 1.4 million figure describes the population Truffle’s initiative targets; it is not a count of confirmed compromises. By connecting verification to issuer-led revocation, the Secrets Revocation Program addresses one concrete part of the API-key security lifecycle. Its effectiveness will depend on issuer participation, sound verification, and a workable path for customers whose credentials are disabled.

![Banner illustrating exposed secrets and credential revocation](https://assets.probackend.com/0cc268be-ea10-4aec-b7bc-1e2306836794?token=ast_7b9ee8975a4b187969d396286e3f9717c42f2331)

### Sources

- [BleepingComputer: Nearly 1.5 million exposed secrets targeted in new revocation program](https://bleepingcomputer.com/news/security/truffle-security-launches-program-to-revoke-14-million-live-secrets)
- [Truffle Security: Secrets Revocation Program](https://trufflesecurity.com/secrets-revocation-program)
