geek vibes.geek vibes.
Tecnología · 6 min · Nov 2, 2026

Trigger.dev Confirmed a Secrets Breach — and the Real Problem Is That Your AI Coding Agent Is Leaking Them Too

Trigger.dev confirmed an attacker accessed customer environment variables. The real story isn't the breach itself: GitGuardian reports AI-service secrets exposure grew 81% this year, and AI-assisted commits leak credentials at twice the rate of the rest.

Ilustración editorial abstracta de un candado digital resquebrajándose en múltiples hilos de luz escapando hacia una red de nodos, en tonos navy, verde esmeralda y cian con acentos violeta.

On Sunday, November 1, Trigger.dev — a cloud background-jobs and automation platform many product teams and agencies use to run workflows, cron jobs, and pipelines — confirmed that an attacker gained unauthorized access to environment variables belonging to a subset of its customers. The company says it has already blocked the exploit, emailed affected account owners with mitigation steps, and will keep posting updates to its status page. On its own, that's just another security incident, the kind that surfaces almost every week.

What makes it relevant today isn't the breach itself — Trigger.dev hasn't said how many organizations were affected or what the exact root cause was, and we're not going to speculate about that — but the fact that it lands literally days after GitGuardian published its annual "State of Secrets Sprawl 2026" report, which documents in hard numbers why this kind of exposure is becoming the norm, not the exception, right as more teams write code with help from AI agents.

What We Know (and Don't) About the Trigger.dev Incident

Confirmed so far, straight from the company's own disclosure:

  • What was exposed: environment variables — the credentials, API keys, and connection strings an app needs at runtime — for a subset of customer organizations.
  • What the company did: blocked the exploit used in the attack, notified affected account owners by email with mitigation instructions, and opened a status page (status.trigger.dev) for ongoing updates.
  • What we still don't know: the exact number of affected organizations and the exploit's root cause. If Trigger.dev publishes a more detailed post-mortem in the coming days, it's worth reading — most incidents like this turn out to be an overly broad access token, a compromised third-party dependency, or an internal endpoint exposed by a misconfiguration.

The Pattern Behind It: Coding Agents Are Accelerating the Secrets Leak

This is where GitGuardian's report stops being just another cybersecurity statistic and becomes relevant to anyone using AI to write or ship code in 2026:

  • 28.6 million new secrets were detected in public GitHub commits during 2025 — a 34% year-over-year increase, and 152% more than in 2021.
  • Of the valid secrets detected in 2022, 64% were still active and exploitable in public repositories four years later. That share climbs with every more recent year: 66%, 70%, and 77% for secrets detected in 2025.
  • And the number that should worry any agency or product team using Claude Code, Copilot, or a similar agent: 1,275,105 AI-service secrets were exposed, an 81% year-over-year jump, and AI-assisted commits leak secrets at twice the rate of the rest of public GitHub commits.

The report also found 24,008 unique secrets exposed specifically in MCP (Model Context Protocol) configuration files — the standard connecting AI agents to external tools — with Google API keys (19.2%) and PostgreSQL connection strings (14%) as the most common types.

CI/CD Pipelines, Not Laptops, Are the New Target

The same report, analyzing the Shai-Hulud 2 supply-chain attack, found something that should change where every technical team points its attention: 59% of compromised machines were CI/CD runners, not developer laptops. And 44% of those compromised machines held more than 10 distinct secrets; 5% held more than 100.

That matters because most teams — agencies included — still treat secrets security as a "don't push your .env to GitHub" problem, when the real risk has already moved to shared infrastructure: the CI/CD runner that deploys five different client projects, the automation account that runs workflows end to end, the Trigger.dev-style service orchestrating jobs on behalf of dozens of organizations.

Why Almost Nobody Rotates Their Secrets, Even Though They Know They Should

GitGuardian also measured something almost no team wants to admit: the rotation rate for exposed secrets sits between 23% and 36%, depending on the year. In other words, the vast majority of leaked credentials never get changed afterward. The report puts it plainly: "Teams can't rotate secrets without risking production outages—so they don't." One more data point connects directly to the Trigger.dev incident: internal (non-public) repositories are 6 times more likely to contain hardcoded secrets than public ones — 32.2% versus 5.6% — precisely because nobody watches them as closely.

What This Means for Agencies, Developers, and Your Product

If your team — in-house or client-facing — uses AI agents to write code, or relies on automation platforms like Trigger.dev, Vercel, GitHub Actions, or similar, this isn't a cybersecurity story that lives outside your day-to-day work:

  • Audit what lives in plain text. Not just in your code: in MCP configuration files, in secrets your AI agents can read or write, and in any automation service running on behalf of a client.
  • Treat secrets rotation as maintenance, not incident response. If your only rotation policy is "rotate when something gets hacked," you're already behind — 70%+ of leaked secrets are still alive years later.
  • Give your shared infrastructure the same scrutiny you give your code. The CI/CD runner or orchestration platform you use to automate delivery across multiple clients is, today, a more attractive target than any single laptop.
  • If you use AI coding agents on client projects, say so in your security proposal. Not to scare anyone, but because GitGuardian's data point (2x more leaks in AI-assisted commits) is exactly the kind of risk a regulated client will ask about sooner or later — better to already have the answer ready.

None of these practices are new. What changed in 2026 is the speed: with AI agents writing and shipping code faster than ever, a bad secrets-handling habit that used to take months to become a problem can now turn into an incident in a single afternoon.

Want us to audit how your team or your automation stack handles secrets and credentials before it becomes a problem? Reach out at hello@geekvibes.agency.

Does any of this resonate?

If you're running pieces that don't quite connect, we'll diagnose it together in 30 min.

→ Schedule a diagnosis
ACTIONS