Non-human identities (service accounts, API keys, tokens, certificates and AI agents) now sit behind a large share of breaches, yet the statistics used to describe them are often untraceable ratios repeated from one slide deck to the next. This page collects non-human identity (NHI) and secrets figures only from the publishers' own report pages, press releases and PDFs: GitGuardian, Verizon, Google Cloud and Mandiant, Microsoft, ENISA, IBM and the Cloud Security Alliance. Every figure carries its report name, publisher and as-of year in the same row, and vendor research is labelled as such. It is written for identity and security leaders, platform and DevOps teams, and anyone who needs to make the case for NHI governance with numbers that stand up to scrutiny. Last reviewed: October 2026.
Key takeaways
- Leaked secrets keep growing faster than developers. GitGuardian's State of Secrets Sprawl 2026 counts 28,649,024 new secrets in public GitHub commits in 2025, with leaks up 152% since 2021 against 98% growth in active developers.
- Leaked secrets stay valid for years. GitGuardian research finds 64% of valid secrets detected in 2022 were still active and exploitable four years later.
- Identity is the main way into cloud. Google Cloud's Threat Horizons Report H1 2026 says identity issues accounted for 83% of Mandiant cloud and SaaS engagements in H2 2025, and stolen credentials of human and non-human identities for 21%.
- Secrets are a common breach payload. The Verizon 2026 Data Breach Investigations Report (DBIR) shows secrets among compromised data in 68% of utilities breaches and 30% of public administration breaches, and credentials in 31% of small and medium business breaches.
- Governance of machine and AI identities is immature. In the Cloud Security Alliance survey sponsored by Oasis Security, 79% rated their confidence in preventing NHI-based attacks as low or moderate and 51% reported no clear ownership.
Headline statistics
| Figure | Report, publisher, as-of year |
|---|---|
| 28,649,024 new secrets detected in public GitHub commits | The State of Secrets Sprawl 2026, GitGuardian (data year 2025) |
| 64% of valid secrets detected in 2022 were still active four years later | The State of Secrets Sprawl 2026, GitGuardian |
| 32.2% of internal repositories contain at least one hardcoded secret, against 5.6% of public repositories | The State of Secrets Sprawl 2026, GitGuardian |
| 1,275,105 AI-service secrets exposed, up 81% year on year | The State of Secrets Sprawl 2026, GitGuardian (data year 2025) |
| 83% of Mandiant cloud and SaaS engagements involved identity issues; 21% began with stolen credentials of human and non-human identities | Cloud Threat Horizons Report H1 2026, Google Cloud (H2 2025 data) |
| Credential abuse was the initial access vector in 13% of breaches, down from first place | Data Breach Investigations Report 2026, Verizon (executive summary) |
| 52.2% of valid account intrusions involved follow-on credential theft | Microsoft Digital Defense Report, Microsoft (report page reviewed October 2026) |
| 24% of organisations take more than 24 hours to rotate or revoke a credential after potential exposure | The State of Non-Human Identity and AI Security, Cloud Security Alliance, sponsored by Oasis Security (383 respondents, August to September 2025, published January 2026) |
Statistics by theme (each with primary source and as-of date)
Secrets leaked in code and developer tooling
| Figure | Report, publisher, as-of year |
|---|---|
| New secrets in public GitHub commits per year: 11 million (2021), 14 million (2022), 18 million (2023), 21 million (2024), 29 million (2025) | The State of Secrets Sprawl 2026, GitGuardian |
| Active GitHub developers rose from 7.6 million (2019) to 22.8 million (2025); 54% made their first commit in 2025 | The State of Secrets Sprawl 2026, GitGuardian |
| Internal repositories are six times more likely than public ones to contain hardcoded secrets (32.2% against 5.6%) | The State of Secrets Sprawl 2026, GitGuardian |
| 46% of critical leaked secrets lack validation checkers, so their validity cannot be tested automatically | The State of Secrets Sprawl 2026, GitGuardian |
| AI-assisted commits leak secrets at twice the baseline rate | The State of Secrets Sprawl 2026, GitGuardian |
| 24,008 unique secrets exposed in Model Context Protocol (MCP) configuration files; top types Google API key 19.2%, PostgreSQL connection string 14%, Firecrawl API key 11.9% | The State of Secrets Sprawl 2026, GitGuardian (data year 2025) |
Public leaks get the headlines, but GitGuardian's internal repository figure is the one most organisations should act on, because private code is where scanning and push protection are least often enforced. Exposure can persist for years: in the Toyota T-Connect Key Exposure 2022, a data server access key sat in a public GitHub repository from December 2017 until September 2022. See the Secrets Management Guide and the Secrets Scanning Buyer's Guide.
Secrets outside code: collaboration tools, CI runners and workstations
| Figure | Report, publisher, as-of year |
|---|---|
| About 28% of secrets sprawl incidents originate from collaboration and productivity tools; 28% of sprawl occurs only outside code repositories and 4% in both | The State of Secrets Sprawl 2026, GitGuardian |
| More than half of collaboration tool incidents are rated critical | The State of Secrets Sprawl 2026, GitGuardian |
| In GitGuardian's analysis of machines compromised by the Shai-Hulud 2 worm, 44% held more than 10 secrets, 5% held more than 100, and 59% were CI/CD runners rather than personal workstations | The State of Secrets Sprawl 2026, GitGuardian |
The CI runner figure explains why worms that steal publishing tokens spread so quickly. The Shai-Hulud npm Worm 2025 took npm tokens, GitHub tokens and cloud credentials from developer laptops and CI runners, posted them to more than 25,000 public repositories and reused the npm tokens to infect more packages. The tj-actions/changed-files Compromise 2025 printed pipeline secrets into workflow logs after a bot account's token was stolen. See the GitHub and GitHub Actions Security Guide.
Remediation and secret lifetime
| Figure | Report, publisher, as-of year |
|---|---|
| Share of valid secrets still active by detection year: 64% (2022), 66% (2023), 70% (2024), 77% (2025) | The State of Secrets Sprawl 2026, GitGuardian |
| Long-lived secrets are the most common policy breach at 60%, followed by internally leaked (17%) and duplicated secrets (16%) | The State of Secrets Sprawl 2026, GitGuardian |
| 24% take more than 24 hours to rotate or revoke a credential after potential exposure; 30% take over a day to triage a high-severity credential leak | The State of Non-Human Identity and AI Security, Cloud Security Alliance, sponsored by Oasis Security (2026) |
| Only 23% of third-party organisations fully remediated missing or improperly secured MFA on cloud accounts; resolving half of weak password and permission misconfiguration findings took almost eight months | Data Breach Investigations Report 2026, Verizon (executive summary, n=7,513) |
The rising "still valid" share for recent years partly reflects less time for rotation; the 2022 figure is the clearer measure of how rarely leaked secrets are revoked. See the Leaked Credential Response Playbook and the Secrets Rotation Runbook Template.
Credentials and secrets in breaches
| Figure | Report, publisher, as-of year |
|---|---|
| Credential abuse as initial access: 13% of all breaches; 25% in EMEA, 20% in Latin America and the Caribbean, 5% in Northern America | Data Breach Investigations Report 2026, Verizon (executive summary) |
| Secrets among compromised data: utilities 68%, agriculture 36%, public administration 30%; by region EMEA 30%, APAC 24%, Northern America 24% | Data Breach Investigations Report 2026, Verizon (executive summary) |
| Credentials among compromised data: management 35%, construction 34%, small and medium businesses 31%, financial 26%, healthcare 25% | Data Breach Investigations Report 2026, Verizon (executive summary) |
| Third-party involvement in 48% of breaches, up 60% on the previous year | Data Breach Investigations Report 2026, Verizon (executive summary) |
| No percentage given, but ENISA describes ShinyHunters using stolen OAuth tokens (ATT&CK T1528) and vishing for SSO credentials, and ransomware operators such as Qilin using stolen or brute-forced VPN credentials, during 2025 | ENISA Threat Landscape 2026, European Union Agency for Cybersecurity |
Verizon's "credentials" category does not separate human passwords from machine secrets, while "secrets" covers keys and tokens. Use both when sizing machine credential exposure. See the Identity Attack Statistics Guide and the Data Breach Statistics Guide.
Cloud identity and service account exposure
| Figure | Report, publisher, as-of year |
|---|---|
| Weak or absent credentials were the initial access vector in 27.2% of Google Cloud incidents in H2 2025, down from 47.1% in H1 2025; third-party software exploitation rose to 44.5% | Cloud Threat Horizons Report H1 2026, Google Cloud (subset of observed activity) |
| Misconfiguration accounted for 21% of Google Cloud initial access in H2 2025 (29.4% in H1 2025); exposed sensitive UI or APIs 4.9% (11.8%) | Cloud Threat Horizons Report H1 2026, Google Cloud |
| Across Mandiant engagements in H2 2025: identity issues 83%, stolen credentials 21%, third-party or trusted relationship compromise 21%, software supply chain compromise 3% | Cloud Threat Horizons Report H1 2026, Google Cloud and Mandiant |
| 78% of observed attack techniques against critical infrastructure used cloud identity abuse | Microsoft Digital Defense Report, Microsoft (report page reviewed October 2026; period not stated) |
| Exposed cloud workloads were attacked after an average of 5.3 hours | Microsoft Digital Defense Report, Microsoft (report page reviewed October 2026) |
Google's report gives examples of stolen GitHub personal access tokens and OAuth tokens but no percentage for service accounts specifically. See the Cloud Workload Identity Guide, the Service Account Security Guide and the Cloud Security Statistics Guide.
NHI and AI identity governance
| Figure | Report, publisher, as-of year |
|---|---|
| 79% rated confidence in preventing NHI-based attacks as low or moderate; 12% were highly confident | The State of Non-Human Identity and AI Security, Cloud Security Alliance, sponsored by Oasis Security (2026) |
| 92% are not confident their legacy IAM tools can manage AI and NHI risks | The State of Non-Human Identity and AI Security, Cloud Security Alliance, sponsored by Oasis Security (2026) |
| 51% report no clear ownership or accountability; 51% name over-permissioned access as a most significant pain point | The State of Non-Human Identity and AI Security, Cloud Security Alliance, sponsored by Oasis Security (2026) |
| 78% lack documented, formally adopted policies for creating or removing AI identities; 14% have fully automated creation and removal, 41% semi-automated, 27% manual | The State of Non-Human Identity and AI Security, Cloud Security Alliance, sponsored by Oasis Security (2026) |
| 97% of organisations with an AI-related breach lacked proper AI access controls | Cost of a Data Breach Report 2025, IBM with Ponemon Institute (600 organisations) |
See the NHI Ownership Guide, the NHI Governance Maturity Model and the AI Security Statistics Guide.
What the numbers mean for defenders
- Detection without revocation does little: GitGuardian's persistence figures and the CSA rotation times point to the same gap. Track time to revoke, not just secrets found, and automate revocation for the providers you use most. See the Cloud Access Key Leak Playbook.
- Scan where secrets actually live: with 28% of sprawl outside code and internal repositories six times more likely to hold secrets, scope scanning to private repositories, CI logs, container images, MCP configs, tickets and chat.
- Replace long-lived secrets: long-lived secrets are 60% of policy breaches in GitGuardian data. Move CI/CD and cloud workloads to federated, short-lived credentials. See the Workload Identity Federation Guide.
- Harden CI runners as identity stores: 59% of Shai-Hulud 2 compromised machines were CI/CD runners. Scope tokens per job, pin actions and remove standing publishing tokens.
- Assign owners: half of CSA respondents report no clear ownership. Every service account, key and agent identity needs a named owner and an expiry. See the Service Account Inventory Template.
- Plan for third-party tokens: Verizon's 48% third-party involvement figure includes integrations holding your tokens. Inventory OAuth grants and vendor-held credentials. See the SaaS API Token Security Guide.
Metrics worth tracking internally
| Metric | External benchmark |
|---|---|
| Median time to revoke a confirmed leaked secret | 24% take more than 24 hours (CSA 2026) |
| Percentage of leaked secrets still valid after 30 days | 64% of 2022 leaks still valid four years later (GitGuardian 2026) |
| Share of private repositories with at least one secret | 32.2% of internal repositories (GitGuardian 2026) |
| Percentage of NHIs with a named owner and expiry date | 51% report no clear ownership (CSA 2026) |
| Share of CI/CD and cloud workloads using federated, short-lived credentials | Long-lived secrets are 60% of policy breaches (GitGuardian 2026) |
Methodology notes and caveats
- GitGuardian research: vendor research from scanning public GitHub commits (GitGuardian states 43% of public commits were scanned) plus customer and incident data. Internal repository and collaboration tool figures come from its customer base, not the whole market.
- Verizon DBIR: sector and region percentages come from the executive summary; small samples in some sectors make single-year shifts unreliable. "Credentials" and "secrets" are separate compromised data varieties.
- Google Cloud Threat Horizons: the Google Cloud initial access table reflects a subset of observed activity in Google Cloud; the Mandiant figures are platform-agnostic engagements. Neither is a market-wide rate.
- CSA survey: self-reported by 383 IT and security professionals; sponsored by Oasis Security, which co-developed the questionnaire.
- Microsoft: figures come from the Digital Defense Report web page, which does not state a period for every figure.
- Machine to human identity ratios: widely quoted ratios (for example from CyberArk's Identity Security Landscape) are omitted because the publisher's own page could not be read for this review. Ratios also depend heavily on what each study counts as a machine identity.
- Refresh cadence: GitGuardian usually publishes in March, Verizon in spring and Google Cloud Threat Horizons twice a year. Check each publisher's page when updating.
Practitioner checklist
- Quote each NHI or secrets figure with report name, publisher, population and year, and label vendor research.
- Inventory service accounts, API keys, tokens, certificates and AI agent identities with owner, purpose and expiry.
- Enable secret scanning and push protection on private repositories, CI logs, container registries and MCP configurations.
- Extend scanning to ticketing, chat and wiki tools where GitGuardian finds over a quarter of sprawl.
- Set and measure a revocation SLA for leaked secrets, with automated revocation for major cloud and SaaS providers.
- Replace long-lived CI/CD and cloud keys with workload identity federation and short-lived tokens.
- Review third-party OAuth grants and vendor-held credentials at least quarterly.
- Map detections to ATT&CK T1552 Unsecured Credentials, T1552.001 Credentials In Files, T1528 Steal Application Access Token and T1078.004 Cloud Accounts.
Standards and references
- GitGuardian: The State of Secrets Sprawl 2026
- Verizon: 2026 Data Breach Investigations Report executive summary
- Google Cloud: Cloud Threat Horizons Report H1 2026
- Cloud Security Alliance: The State of Non-Human Identity and AI Security survey findings
- ENISA: Threat Landscape 2026
- Microsoft: Digital Defense Report
- IBM Newsroom: Cost of a Data Breach Report 2025 findings
- MITRE ATT&CK: Unsecured Credentials
Related NHI Mgmt Group resources: Non-Human Identity (NHI) Guide · Top 10 NHI Risks · Cybersecurity Statistics Guide · Supply Chain Attack Statistics Guide · Board Briefing: NHI Risk