Expanding Credential Layers Blind Enterprise Security Teams

Key takeaways

• Hardcoded secrets in public repositories rose 34% to 28.65 million in 2025, with AI-related credential leaks surging 81%.
• Monthly GitHub commits hit 2.9 billion in August 2026, forcing platform engineering to plan capacity for 30x scale to accommodate agentic AI workflows.
• Developer workstations serve as primary attack targets: 44% of systems compromised by Shai-Hulud 2 held over 10 unique secrets.
• Active credentials persist unchecked across environments, with 64% of secrets verified valid in 2022 still active in 2026.
• Adversary eCrime breakout time dropped to 29 minutes, with lateral movement observed in 27 seconds, requiring continuous discovery before prevention.

Enterprise credential security is failing because software generation is expanding far beyond the boundaries of sanctioned vaults and repository scanners. Today, human engineers and autonomous coding agents scatter authentication keys across workstations, internal code repositories, team collaboration tools, and local configuration files. Attackers exploit this distributed credential layer to establish initial access and move laterally in minutes. Without establishing visibility across unmanaged endpoints and decentralized repositories to calculate an accurate total inventory of active secrets, security organizations cannot effectively remediate exposures or prevent unauthorized access. Defense requires discovering every existing token, mapping its context and permissions, and closing the gap between managed vaults and shadow credentials.

The breakdown of traditional secrets management

Modern enterprise architecture relies entirely on credentials to connect humans, microservices, cloud infrastructure, and autonomous software. Traditionally, security leaders assumed that sensitive tokens lived primarily inside sanctioned vaults or were introduced into code repositories during development. Consequently, enterprises deployed central secrets management systems and periodic repository scanning to police exposure. However, the volume and velocity of code creation have shattered the operational assumptions underlying these legacy controls.

Platform telemetry highlights the unprecedented pace of code generation. According to reporting by The New Stack, GitHub monthly commit volume escalated from approximately 1 billion total commits across all of 2025 to 2.9 billion commits in August 2026 alone. This represents an annualized pace exceeding 14 billion commits, prompting GitHub engineering to redesign capacity planning for 30 times its current scale to accommodate the surge driven by agentic artificial intelligence workflows. Every incremental commit, automated deployment pipeline, and third-party integration introduces new interfaces that require persistent machine-to-machine authentication.

As software production accelerates, credential leakage increases exponentially. In its State of Secrets Sprawl 2026 report, GitGuardian revealed that it detected 28.65 million new hardcoded secrets in public GitHub commits during 2025—a 34% increase year over year. Over the multi-year study period, the rate of secret exposure outpaced the growth of active software developers by a factor of 1.6. Crucially, leaked credentials associated specifically with artificial intelligence services surged by 81%, reflecting how rapidly unvetted integrations enter development pipelines.

The primary structural challenge for enterprise security leaders is that the credential layer lacks a defined physical or network perimeter. A secret generated in an approved cloud management console frequently migrates into unmanaged environments: a developer pastes an API key into a staging configuration, saves an access token within a local shell script, or mirrors an environment variable inside an internal repository. GitGuardian found that internal enterprise repositories are approximately six times more likely than public repositories to contain at least one hardcoded secret. Furthermore, roughly 28% of secret exposure incidents originate completely outside source code systems, proliferating through ticketing queues, shared productivity wikis, and enterprise messaging platforms.

Developer endpoints as supply chain beachheads

Because software engineers build, test, and debug distributed applications locally, developer workstations have accumulated dense concentrations of unmanaged authentication material. For years, development practices normalized storing credentials in local environment configuration files, caching cloud session tokens in command-line tool directories, and leaving plain authentication strings in unscrubbed terminal shell histories. Organizations viewed the risk of local storage as relatively low because developer machines sat behind corporate firewalls and endpoint detection agents.

Adversaries have systematically capitalized on this security blind spot. Information-stealing malware campaigns that emerged in late 2025 and early 2026, including Shai-Hulud and S1ingularity, transformed developer endpoints into high-yield targets for supply chain infiltration. As detailed in analysis published by The Hacker News, GitGuardian evaluated 6,943 developer workstations and environments compromised during the Shai-Hulud 2 supply chain campaign. Researchers uncovered 33,185 unique secrets across those machines. Approximately 44% of compromised workstations contained more than 10 unique credentials, while 5% harbored over 100 distinct secrets.

When an information stealer compromises a developer workstation, the malware does not respect organizational boundaries or tool silos. The malware programmatically scours browser profiles, terminal histories, local application caches, and configuration directories to extract all available authentication keys. Because a single developer laptop often possesses access to corporate identity providers, production cloud infrastructure, internal Git repositories, and third-party SaaS APIs, compromising that single endpoint grants adversaries lateral access across segregated enterprise domains. Organizations seeking to protect their broader software ecosystems must examine how unmanaged developer tokens intersect with wider software supply chain security risks.

Agentic AI and the amplification of authentication risk

The concentration of credentials on developer endpoints is compounded by the rapid introduction of autonomous coding agents into daily engineering workflows. Unlike conventional static analysis tools, autonomous AI agents operate with direct execution privileges: they inspect file trees, execute terminal commands, orchestrate local test suites, and interact dynamically with remote services. As enterprises explore agentic security governance and internal risk controls, managing these agent connections has emerged as a frontline operational challenge.

A primary integration mechanism facilitating agent autonomy is the Model Context Protocol (MCP), which enables coding agents to interface directly with external databases, software repositories, and enterprise APIs. However, authenticating these agentic workflows has opened an unmonitored attack vector. In 2025, GitGuardian identified 24,008 unique secrets exposed in publicly accessible Model Context Protocol configuration files. Verification testing confirmed that 2,117 of those credentials were fully active and valid, granting direct access to backend systems and services.

Get the Weekly Brief

Curated analysis for tech leaders. Every Thursday.

Subscribe

The risk of agentic credential exposure extends beyond deliberate adversary theft. When autonomous agents operate with broad, unvetted authentication tokens, unexpected programmatic behaviors or hallucinated logic can trigger destructive actions across production databases and cloud environments. Restricting agent privileges requires granular identity scoping, yet enterprise security teams frequently lack visibility into the local configuration files and temporary tokens that engineers configure to keep their AI assistants operational.

The denominator problem in coverage metrics

Most enterprise security leaders operate under a false sense of security regarding secrets management. Organizations frequently point to centralized vault implementations containing tens of thousands of securely managed credentials as evidence of compliance and security maturity. However, measuring coverage exclusively by counting vaulted secrets creates a dangerous “denominator problem.”

If an enterprise maintains 50,000 secrets within an approved vault, that metric only describes the credentials security teams already know about. It reveals nothing about the thousands of unmanaged, plaintext credentials scattered across decentralized Git repositories, local developer workstations, and collaboration tools. When calculating secrets coverage, the true denominator is the entire population of active credentials across the enterprise ecosystem. Without automated discovery across all operational environments, security teams remain blind to shadow credentials operating entirely outside sanctioned governance paths.

The severity of this visibility gap is worsened by the extraordinary longevity of leaked credentials. Security models often assume that exposed tokens naturally expire or become obsolete over time. However, GitGuardian retested credentials that were verified as valid in 2022 and found that 64% remained active in January 2026. A token abandoned in an archived repository or an unmanaged shell history can remain usable to adversaries for four years or longer.

Adversaries recognize this reality and exploit it ruthlessly. According to findings from the 2026 Verizon Data Breach Investigations Report, compromised credentials represented 22% of initial access vectors, driven significantly by corporate credentials discovered on unmanaged endpoints. Once attackers obtain valid credentials, defense response windows collapse. The 2026 CrowdStrike Global Threat Report recorded that average eCrime breakout times dropped to 29 minutes in 2025—a 65% acceleration compared to 2024—with the fastest observed adversary lateral movement occurring in just 27 seconds. Human defenders cannot intercept attacks operating at machine speed; security strategies must pivot from reactive triage to systematic discovery and continuous prevention.

Bottom line

Securing the enterprise credential layer requires establishing complete visibility before attempting remediation or policy enforcement. Organizations cannot prevent leaks or revoke risky credentials when they only monitor a fraction of their actual token population. To build an accurate credential inventory, security teams must deploy continuous discovery across source code repositories, collaboration tools, and developer endpoints.

Every discovered credential must be enriched with critical context: active validity status, complete exposure history through credential fingerprinting, workload ownership, privilege scope, and infrastructure dependencies. By identifying the true denominator of all active tokens, enterprise security leaders can eliminate blind spots, protect developer environments from infostealer malware, establish guardrails around autonomous AI agents, and transition their secrets management from reactive cleanup to automated prevention.

Sources

• The State of Secrets Sprawl 2026 | GitGuardian Annual Report
• GitHub now sees 2.9 billion commits a month – and it can’t keep up – The New Stack
• The Credential Layer Is Expanding Faster Than Security Teams Can See It
• Initial Access Changed, The Attack Path Did Not: Findings From The Verizon 2026 DBIR
• 2026 CrowdStrike Global Threat Report: AI Accelerated Adversaries