26,400 Live Cloud and AI API Keys Found in Public Gitee Repositories — What Enterprises Must Do Now
Executive Summary
In October 2026, independent researcher Ben Zimmermann published findings from a month-long scan of 7.25 million public Gitee repositories, uncovering 26,400 valid, live credentials for cloud and AI services. Using TruffleHog and Kingfisher scanners against full commit histories, he confirmed active keys for Alibaba Cloud, DeepSeek, OpenAI, YouTube API, Zhipu BigModel, AWS, GitHub, and Google Cloud. One credential was an active organization-admin key for a major enterprise AI platform. Vendors were notified, but no confirmation of revocation was received before publication, leaving any organization with keys in public Gitee history at immediate risk of account takeover, quota abuse, unexpected charges, and unauthorized data access.
This incident underscores a dangerous acceleration in AI-era credential sprawl: as developers rapidly integrate cloud and AI APIs, secret hygiene practices are not keeping pace. With nearly half of exposed credentials dated to 2025 alone, the attack surface is expanding in real time, and a single leaked organization-admin key can grant adversaries deep access to enterprise AI infrastructure.
Why This Matters Now
AI API key adoption is surging globally in 2026, and Gitee hosts millions of repositories spanning enterprise, academic, and government contributors. With 3,170 DeepSeek and 1,348 OpenAI keys confirmed live—and no systematic revocation confirmed at publication—any threat actor can harvest working credentials today. Regulatory frameworks including PCI DSS 4.0 and NIST CSF now mandate secrets management controls, making this a compliance liability as well as an operational risk.
Attack Path Analysis
Attackers and opportunistic actors harvested 26,400 valid cloud and AI credentials (Alibaba Cloud, OpenAI, DeepSeek, AWS, Google Cloud, GitHub) exposed in public Gitee repositories, where developers had inadvertently committed secrets into version history. With valid, long-lived API keys in hand, adversaries could directly authenticate to target cloud and AI services without any exploit, escalate to broader resource access using permissive IAM roles tied to those keys, pivot laterally across cloud accounts and AI platforms, issue API-driven commands or exfiltrate data through legitimate service endpoints, and ultimately cause financial loss, quota exhaustion, data theft, or destruction of cloud resources.
Kill Chain Progression
This analysis maps confirmed threat intelligence to the full cloud kill chain to show where defensive gaps would emerge as an attack progresses.
Initial Compromise
Description
Threat actors or automated scanners harvested 26,400 valid cloud and AI API credentials (Alibaba Cloud AccessKey pairs, OpenAI keys, DeepSeek keys, AWS keys, Google Cloud/YouTube API keys, GitHub tokens) that developers had committed to public Gitee repositories, including historical commit objects never purged. Tools such as TruffleHog and Kingfisher were used to scan 7.25 million repositories and validate each credential live against the issuing service. No exploit was required; the credentials were openly accessible via Gitee's public API and git clone. One credential was confirmed as an active organization-admin key for a major enterprise AI platform.
MITRE ATT&CK® Techniques
Unsecured Credentials: Credentials In Files
Data from Information Repositories
Valid Accounts: Cloud Accounts
Data from Cloud Storage
Impair Defenses: Disable or Modify Cloud Logs
Resource Hijacking
Data Destruction
Cloud Service Discovery
Potential Compliance Exposure
Mapping incident impact across multiple compliance frameworks.
PCI DSS 4.0 – Protection of Stored Credentials and Authentication Management
Control ID: Requirement 3.5 / 8.3
NYDFS 23 NYCRR 500 – Access Privilege Management and Training on Cybersecurity Awareness
Control ID: 500.7 / 500.14
DORA (Digital Operational Resilience Act) – ICT Security and Access Control / Detection Capabilities
Control ID: Article 9 / Article 10
CISA Zero Trust Maturity Model 2.0 – Credential Hygiene and Least-Privilege Access Enforcement
Control ID: Identity Pillar — Credential Management / Least Privilege
NIS2 Directive – Access Control Policies and Incident Handling for Essential/Important Entities
Control ID: Article 21 — Security of Network and Information Systems
ISO/IEC 27001:2022 – Data Leakage Prevention, Information Backup and Identity Management
Control ID: Annex A 8.12 / 8.13 / 5.16
Sector Implications
Industry-specific impact of the vulnerabilities, including operational, regulatory, and cloud security risks.
Information Technology/IT
Exposure of 26,400 live cloud and AI credentials in public Gitee repositories creates direct risk of unauthorized access, quota exhaustion, lateral movement, and data exfiltration across multi-cloud and hybrid IT environments.
Computer Software/Engineering
Developers committing Alibaba Cloud, OpenAI, and AWS keys into public repositories trigger credential exposure incidents exploitable for privilege escalation and egress policy violations per NIST.800-53.AC-6 and PCI.4.0 controls.
Financial Services
Exposed organization-admin AI platform credentials and cloud access keys risk unauthorized financial data access, unexpected service charges, and compliance failures under HIPAA, PCI 4.0, and NIST Zero Trust mandates.
Health Care / Life Sciences
Leaked cloud credentials with unrevoked API access threaten protected health data stored across multi-cloud workloads, violating HIPAA.164.312(e)(1) and ZTMM data controls requiring encrypted transit and egress enforcement.
Sources
- 26,400 Valid Cloud and AI Credentials Found Exposed in Public Gitee Repositorieshttps://thehackernews.com/2026/10/26400-valid-cloud-and-ai-credentials.htmlVerified
- Scanning 7.25 Million Public Repositories on China's Largest Git Platform for Secretshttps://www.mongodb.com/company/blog/technical/scanning-7-point-25-million-public-repositories-chinas-largest-git-platform-for-secretsVerified
- Why Exposed Credentials Stay Live for Yearshttps://trufflesecurity.com/blog/why-exposed-credentials-stay-live-for-yearsVerified
- TruffleHog - Open Source Secret Scannerhttps://github.com/trufflesecurity/trufflehogVerified
- Kingfisher - MongoDB Open Source Secret Scannerhttps://github.com/mongodb/kingfisherVerified
- Alibaba Cloud AccessKey Leakage Response Planhttps://help.aliyun.com/en/ram/user-guide/solution-to-accesskey-leakageVerified
- OpenAI Best Practices for API Key Safetyhttps://help.openai.com/en/articles/5112595-best-practices-for-api-key-safetyVerified
- Google Cloud API Key Authentication and Rotationhttps://docs.cloud.google.com/docs/authentication/api-keysVerified
- Scanning 5.6 Million Public GitLab Repositories for Secretshttps://trufflesecurity.com/blog/scanning-5-6-million-public-gitlab-repositories-for-secretsVerified
Frequently Asked Questions
Cloud Native Security Fabric Mitigations and ControlsCNSF
Based on the attack progression modeled above, these are the defensive controls that would constrain each stage.
Aviatrix Zero Trust CNSF is directly relevant to this incident because exposed cloud credentials harvested from public Gitee repositories gave adversaries authenticated access to cloud management planes, AI APIs, and storage services across multiple providers. By enforcing segmented workload access, identity-scoped routing, and controlled egress paths, CNSF would likely reduce the blast radius available to an actor operating with stolen but valid credentials.
Control: Cloud Native Security Fabric (CNSF)
Mitigation: Even with valid credentials in hand, CNSF's fabric-wide visibility and policy enforcement would likely constrain which cloud management plane endpoints and workloads an actor could reach, reducing the usable attack surface tied to each exposed key.
Control: Zero Trust Segmentation
Mitigation: Zero Trust segmentation would likely constrain an actor's ability to move from one identity boundary to another, reducing the privilege scope reachable through role assumption or token exchange by ensuring that only explicitly permitted identity-to-resource paths are allowed.
Control: East-West Traffic Security
Mitigation: East-west traffic enforcement would likely reduce the reachability of cloud services and workloads that an actor attempts to pivot into using stolen credentials, constraining cross-account and cross-service movement to only those paths explicitly permitted under policy.
Control: Multicloud Visibility & Control
Mitigation: Multicloud visibility across the fabric would likely surface unusual API call patterns or unexpected management plane traffic originating from workloads outside their normal operational profile, enabling policy-based responses that could constrain an actor's ability to maintain persistent, undetected API-driven control.
Control: Egress Security & Policy Enforcement
Mitigation: Egress security and policy enforcement would likely constrain the outbound paths available to an actor attempting to transfer data to external destinations, reducing the volume and reach of data that could be moved outside controlled cloud boundaries using stolen credentials.
Where prior CNSF controls constrained lateral movement, privilege escalation, and egress paths, the residual impact would likely be scoped to the specific services and accounts directly accessible under the stolen credential's permitted identity, rather than spreading across the full multicloud estate.
Impact at a Glance
Affected Business Functions
- Cloud Infrastructure Access Management (Alibaba Cloud, AWS, Google Cloud)
- AI and LLM API Service Consumption (OpenAI, DeepSeek, Zhipu BigModel, Gemini)
- Developer Platform and Source Code Repository Operations
- Enterprise AI Platform Administration (confirmed organization-admin credential exposure)
Estimated downtime: N/A
Estimated loss: N/A
26,400 confirmed valid credentials were found exposed across 7.25 million public Gitee repositories. Credential categories include: 9,473 Alibaba Cloud AccessKey pairs enabling full cloud resource control; 3,805 YouTube/Google Cloud API keys potentially granting unrestricted access to all Google Cloud APIs; 3,170 DeepSeek AI API keys; 1,489 Zhipu BigModel AI platform keys; 1,348 OpenAI API keys enabling unauthorized quota consumption and potential data access; smaller counts of GitHub, AWS, and Google Cloud credentials. At least one credential was confirmed to be an active organization-admin credential for a major enterprise AI platform, representing a severe privilege escalation risk. Exposed credentials span commit history dating from the late 2000s through 2025, with 12,242 originating in 2025 alone. No confirmed revocations had occurred at time of publication, meaning all 26,400 credentials remained actionable for unauthorized use, account takeover, unexpected financial charges, data exfiltration, and cloud resource manipulation.
Recommended Actions
Key Takeaways & Next Steps
- • Enforce egress security and policy enforcement (CNSF Egress Security & Policy Enforcement) to detect and block outbound API calls from cloud workloads to unauthorized or unexpected external AI and cloud service endpoints, preventing stolen credentials from being leveraged undetected through legitimate HTTPS channels.
- • Deploy multicloud visibility and control (CNSF Multicloud Visibility & Control) to establish behavioral baselines for API usage across Alibaba Cloud, AWS, Google Cloud, and AI platforms, enabling detection of anomalous authentication patterns, unusual resource access volumes, or API calls from unexpected geolocations indicative of credential abuse.
- • Apply zero trust segmentation (CNSF Zero Trust Segmentation) with identity-based, least-privilege policies so that even if a developer API key is compromised, its blast radius is contained—keys should be scoped to specific workloads, namespaces, or SmartGroups and denied access to cross-service or cross-account resources by default.
- • Leverage cloud firewall capabilities (CNSF Cloud Firewall / ACF) with FQDN and URL filtering to restrict which cloud management plane endpoints workloads and CI/CD pipelines are permitted to call, reducing the attack surface available to an adversary using harvested API credentials and limiting exfiltration paths to cloud storage or AI APIs.
- • Utilize inline IPS and threat detection (CNSF Inline IPS / Suricata and Threat Detection & Anomaly Response) to identify signature patterns associated with secret-scanning tooling (TruffleHog, Kingfisher) and automated credential validation traffic, and to alert on bulk API enumeration or resource creation bursts that indicate active abuse of exposed cloud or AI keys.
