The Credential Layer Is Expanding Faster Than Security Teams Can See It

Security teams gain a clear view of where secrets actually live once they map the credential layer across repositories, endpoints, and collaboration tools, rather than relying on vault counts alone. The consequences of missing that map are growing fast: secret exposure is rising faster than the developer population, and the fastest observed attacker breakout now sits at 27 seconds, which makes the case for discovery before remediation or prevention.

Why software production is outpacing secret management

Commit volume on GitHub jumped from roughly 1 billion across all of 2025 to 2.9 billion in August 2026, putting the platform on pace for more than 14 billion commits in the 2026 reporting year. GitHub’s own engineering team has moved from planning for 10x scale to designing for 30x scale as agentic development accelerates. Every additional application, integration, and agent is another system that needs to authenticate, which is another credential that has to be tracked.

The exposure numbers confirm the trend. 28.65 million new hardcoded secrets were detected in public GitHub commits in 2025, up 34% year over year, and credentials linked to AI services grew 81%. Over the full research window, secret exposure rose 1.6 times faster than the count of active developers.

Where the credential layer actually lives

The credential layer includes every secret that connects people, applications, infrastructure, and services, and it crosses every technology and ownership boundary a security team might draw.

Several pockets of that layer stay invisible to standard tools:

• Internal repositories were roughly six times more likely than public repositories to contain at least one secret, according to GitGuardian’s State of Secrets Sprawl 2026.
• Collaboration and productivity systems were the origin of about 28% of secrets incidents, with credentials pasted into tickets, sent through chat, and left searchable long after the work ended.
• Developer laptops hold local environment files, cached cloud credentials, and shell history that an infostealer can sweep up in seconds.
• AI agent configuration introduces a newer pocket, with Model Context Protocol connections and local agent files creating new places where authentication material gets stored.

Each scanner, vault, repository, and endpoint holds a separate slice of credentials, and no single tool sees the combined population until a discovery layer stitches those slices together.

Attackers already search across those boundaries

Compromised credentials accounted for 22% of initial access in the 2026 Verizon Data Breach Investigations Report, and corporate credentials on unmanaged devices drove a significant share of the breaches in that dataset. The reporting period closed before the Shai-Hulud and S1ingularity infostealer worms spread widely in early 2026, so the real share is now likely higher.

Self-propagating malware running on an endpoint can pull authentication material from browser data, local files, and application storage without caring which team created it. A developer machine is an especially rich target because it can authenticate to source control, cloud infrastructure, and local applications under development. Compromising one endpoint can expose access across otherwise separate parts of the enterprise.

What the developer laptop is now holding

Two shifts turned the developer endpoint into a first-class part of the credential layer. The first is the wave of infostealer campaigns that target developer machines directly. The second is the rise of AI coding agents that share the same machine and use the same credentials.

Analysis of systems compromised during the Shai-Hulud 2 supply-chain campaign found 33,185 unique secrets across 6,943 compromised machines. Forty-four percent of those machines held more than 10 secrets, and 5% held more than 100. The same research window turned up 24,008 unique secrets in public MCP configuration files during 2025, of which 2,117 were verified as still valid.

A coding agent that has access to production credentials can take harmful actions without oversight.

Why context matters for every credential found

Identifying a secret is the first step. The surrounding context, including nearby code and comments, determines how a detected secret should be handled.

• Validity is the most immediate signal. GitGuardian retested credentials confirmed valid in 2022 and found 64% still valid in January 2026, so a leaked secret can stay useful to an attacker for years.
• Location history shows how far a credential has traveled. A secret that has appeared on a laptop and in a public repository has crossed more boundaries than one found once in an internal repo.
• Credential fingerprinting connects those appearances to a single record, so seven detections of the same secret count as one credential with seven known exposures.
• Ownership links the finding to the user, workload, or application that relies on it, including the policy and governance that apply.
• Permission scope separates a development credential from a production credential with broad administrative access.
• Dependencies show what workloads a credential supports, which is the information needed to rotate or revoke it safely.

Vault coverage is one number, not the whole map

Many enterprises already run strong secrets-management programs, and a vault inventory is a useful starting point. It only covers the credentials that were placed inside it. GitGuardian’s research points to organizations holding 50,000 credentials in approved vaults while thousands more sit in plaintext across repositories, endpoints, and collaboration tools, with no link between the two populations.

Useful coverage questions start with the full credential population: how many active credentials can be tied to permissions and dependent workloads, how many have crossed into public environments, and how many have an identified owner at all. Those answers depend on discovery first.

Why the response window is shrinking

CrowdStrike reported an average eCrime breakout time of 29 minutes in 2025, with the fastest observed case reaching lateral movement in 27 seconds. A human-speed response cannot keep up, and waiting to discover credentials until a periodic scan runs only widens the gap. The path forward is discovery that produces a context-rich inventory, then remediation and prevention built on top of that inventory.

Rank position across a service area is what a geo grid report shows, and the same idea applies here: a single city-wide average cannot tell a security team which parts of the credential layer are invisible.

FAQ

How fast is secret sprawl really growing?

GitGuardian detected 28.65 million new hardcoded secrets in public GitHub commits in 2025, up 34% year over year, and credentials linked to AI services rose 81%. Over the full research window, secret exposure grew 1.6 times faster than the active developer count.

Why are developer laptops now part of the credential layer?

Infostealer campaigns such as Shai-Hulud and S1ingularity turned developer endpoints into a primary target. Analysis of Shai-Hulud 2 compromises found 33,185 unique secrets across 6,943 machines, with 44% of compromised machines holding more than 10 secrets and 5% holding more than 100.

What context should discovery attach to each credential?

A useful record includes validity, location history, fingerprinting across appearances, ownership, permission scope, and dependencies on other workloads. Validity matters because 64% of credentials confirmed valid in 2022 were still valid in January 2026.