Truffle Security launches revocation program for 1.4 million live exposed secrets

Key facts

The 1.4 million secret backlog and issuer-side revocation

On Wednesday 4 November 2026, cybersecurity research firm Truffle Security announced the launch of the Secrets Revocation Program, an industry initiative offering credential-issuing platforms free, automated mechanisms to invalidate publicly leaked API keys. The program launched alongside the disclosure of 1.4 million verified, active credentials discovered across public code repositories, container images, package registries, and artificial intelligence datasets.

Authored by researchers Luke Marshall and Dylan Ayrey, the disclosure marks a strategic shift from passive developer notification to automated issuer-side revocation. Truffle Security demonstrated that emailing developers regarding exposed keys is largely ineffective: across 21,981 live credentials reported to developers on GitLab and Bitbucket in prior investigations, exactly 81% (17,805 keys, derived) remained functional six months later. In public repository snapshots, unmanaged credentials persist indefinitely, with 543,699 active keys identified in The Stack v3 archive having remained exposed for over two years.

Zoom

Operational metrics and scale of the Truffle Security Secrets Revocation Program

Operational Domain	Recorded Metric	Measurement Standard	Analytical Significance
Live Secret Backlog	1.4 Million Keys	Active credentials confirmed functional	Immediate global exposure across public web
Scanning Throughput	>20 Petabytes (2026)	6 PB scanned in October 2026 alone	Comprehensive multi-registry inspection
Discovery Rate	1 Key / 150 Seconds	24 keys/hour (576 keys/day, derived)	Continuous inbound credential leakage velocity
Notification Failure	81% Active (6 Months)	17,805 of 21,981 reported keys (derived)	Empirical failure of passive developer notification
Multi-Year Debt	543,699 Active Keys	The Stack v3 repository archive	Secrets remain functional indefinitely unless revoked
Partner Ecosystem	16 Launch Issuers	43,699 verified keys queued in backlog	Hugging Face, Vercel, Notion, Resend, Twilio

Truffle Security emphasized that the program operates entirely free of charge for any platform that issues API keys or authentication tokens. By establishing direct API integrations and cryptographically signed webhook notifications, the program enables service providers to invalidate leaked customer credentials immediately upon discovery, neutralizing exposure before automated adversarial scrapers can exploit them.

Empirical comparison: passive notification versus active revocation

For more than a decade, the standard cybersecurity industry response to exposed credentials has been advisory notification: scanning systems detect a secret and transmit an automated email alert to the committer. Truffle Security presented empirical telemetry demonstrating that passive notification fails to protect enterprise infrastructure. In a benchmark study tracking 21,981 verified credentials reported directly to developers on GitLab and Bitbucket, 81% of the keys remained active six months later.

Zoom

Survival rates of leaked credentials across platform revocation policies

Credential Platform	Revocation Policy Model	Active Leak Proportion	Operational Outcome
Google Cloud (Pre-June 2024)	Customer email notification only	96% of 69,041 active keys (derived)	Keys remained live indefinitely across repositories
Google Cloud (Post-June 2024)	Automated default service account disabling	Sharp drop to near-zero active leaks	Demonstrates efficacy of issuer-side intervention
Hugging Face / OpenAI / Anthropic	Immediate automatic key revocation	<= 2.2% active keys in The Stack v3	Exposed keys neutralized before exploitation
Telegram Bot Tokens	Zero reporting channel / no revocation	53% active keys in The Stack v3	Persistent open exposure across open-source code

The effectiveness of automated revocation is validated by Google Cloud's policy transition. Prior to 16 June 2024, Google notified customers via email when service account keys leaked to GitHub. On that date, Google shifted to automatically disabling leaked service account keys by default. Analyzing Google Cloud credentials in The Stack v3, Truffle Security found that of 69,041 working GCP keys, exactly 96% (66,279 keys, derived) were committed prior to the policy change, confirming that automated disabling permanently alters key survival curves.

Non-GitHub blindspots: Docker Hub and Hugging Face

While developer security tools have focused heavily on GitHub source code repositories, Truffle Security's research revealed that massive volumes of active credentials reside outside traditional source code scanners. GitHub's native partner scanning program covers public GitHub repositories and npm packages, leaving public container registries and machine learning repositories uninspected.

Zoom

Distribution of exposed secrets across modern deployment and AI repositories

Platform / Registry	Observed Exposure Metric	Primary Root Cause	Scanner Visibility Blindspot
Docker Hub Containers	75% of exposed Resend API keys (derived)	Secrets baked into container build layers	Invisible to native Git source code scanners
Hugging Face Datasets	>50% of exposed Vercel API tokens	Unsanitized machine learning training corpora	Bypasses standard CI/CD pre-commit hooks
AI Local Config Files	2,655 working keys (.claude/settings.local.json)	Local assistant settings packaged into Docker	Completely unobserved prior to 2025 (derived)
Agentic Pull Requests	6.8% of key-leaking commits signed by AI	AI agents writing 40% of GitHub PRs	Accelerates automated introduction of raw credentials

The container blindspot is particularly severe. Three out of every four (75%, derived) exposed Resend API keys delivered to the issuer originated from public Docker container images rather than source repositories. Similarly, over 50% of exposed Vercel API tokens identified in the analysis were discovered inside public datasets hosted on Hugging Face, illustrating how machine learning training corpora inadvertently capture and redistribute production API keys.

The AI multiplier: agentic code generation and config leaks

The rapid adoption of artificial intelligence in software engineering has accelerated both credential leakage and adversarial exploitation. Telemetry cited by Truffle Security reveals that pull requests originating as an AI agent draft rose from 17% in March 2026 to over 40% in September 2026 (a 23 percentage point increase, derived). Furthermore, while zero percent of commits leaking working keys were signed by AI agents in mid-2025, that proportion reached 6.8% by May 2026.

Local AI coding tools have introduced a dangerous shadow credential vector. Developers running local AI assistants frequently maintain environment variables, session cookies, and API tokens inside project dot-directories. When developers package their workspace into a Docker container, hidden files such as .claude/settings.local.json are committed into container layers. Truffle Security has identified 2,655 functional credentials in AI coding tool configuration files since January 2025, compared to exactly zero in 2024.

Adversaries exploit these exposed tokens to power automated attacks. In July 2026, researchers demonstrated that 14 Hugging Face tokens discovered in a public dataset enabled autonomous agents to compromise production infrastructure within 24 hours. Furthermore, threat actors systematically steal foundation model API keys to conduct unauthorized model distillation, querying commercial models millions of times at the victim's expense to train competing models.

Program architecture and issuer integration

The Secrets Revocation Program establishes a standardized pipeline bridging discovery and automated remediation. First, TruffleHog actively validates every discovered secret against live provider endpoints to confirm authenticity, filtering out stale, expired, or synthetic test strings. Second, verified live keys are dispatched to participating credential issuers via cryptographically signed webhooks or dedicated revocation APIs. Third, the issuer terminates the token immediately.

Sixteen technology companies joined the initiative at launch, including Hugging Face, Vercel, Notion, Resend, RubyGems, and Twilio. Truffle Security is currently holding 43,699 verified, functional credentials in its delivery pipeline awaiting partner onboarding by unjoined API providers. The firm urged all organizations that issue developer API keys to integrate with the free program, emphasizing that in an era of machine-speed AI attacks, passive developer notification cannot protect cloud infrastructure.