Click to upload or drag and drop
PDF, DOCX, PPTX, TXT, JPG, JPEG, PNG, HEIC, ODP, ODT, BMP, or TIFF
up to 20MB
Uploading...
GitHub certification changes in 2026 touched the entire exam family: GH-900 Foundations and GH-200 Actions changed significantly in January 2026, GH-100 Administration and GH-500 Advanced Security changed significantly in July 2026, and GH-600 Developing in Agentic AI Systems launched as a brand-new beta exam in the spring. Those are the official change logs' own words on Microsoft Learn, not our paraphrase. The only quiet corner is GH-300 Copilot, whose latest change log marks every functional group as No change. Practical consequence: almost every course, book and question bank for GitHub certs written before 2026 now tests a previous exam.
Here is the exam-by-exam record, from the official study guides and certification pages, all read directly in July 2026.
| Exam | What happened in 2026 | Headline additions |
|---|---|---|
| GH-900 Foundations | Changed significantly, January 2026 | Copilot agents, Agent Mode, multi-model support, passkeys, Enterprise Managed Users, github.dev |
| GH-200 Actions | Changed significantly, January 2026 | Immutable actions, OIDC federation, artifact attestations with SLSA provenance, YAML anchors, Ubuntu 20.04 runner deprecation |
| GH-300 Copilot | Change log: every functional group No change | Outline already covers Agent Mode, MCP, Copilot CLI, Spaces and Spark |
| GH-100 Administration | Changed significantly, July 2026 | GHEC with Data Residency plus EMU, Azure private networking for runners, third-party vaults, metered product optimization |
| GH-500 Advanced Security | Changed significantly, July 2026 | Renamed domains (Secret Protection, supply chain security, Code Security), EPSS scoring, security campaigns, delegated bypass, autofix |
| GH-600 Agentic AI | New exam, in beta as of July 2026 | MCP servers and registries, multi-agent orchestration, agent memory, guardrails; 120 minutes vs the family's 100 |
One quotable way to hold the whole year in your head: of the six GitHub certification exams that exist in 2026, four were rewritten significantly within seven months, one launched from scratch, and exactly one was left untouched. No other major vendor certification family turned over that fast this year. If your prep material predates the change log for your exam, it is describing a platform that no longer matches the paper, and the two July rewrites mean this is true even for material bought this spring.
The GH-900 rewrite pushed AI into a fundamentals exam: the current outline names Copilot agents, Agent Mode and multi-model support on a test aimed partly at project managers and students, alongside modern account security in the form of passkeys and Enterprise Managed Users. The GitHub Foundations practice test page maps the seven domains and why describe-level verbs dominate. GH-200 got more technical at the edges: the rewritten outline names immutable actions, OIDC federation and SLSA attestations, which is supply-chain security arriving in the CI/CD exam; the domain math on the GitHub Actions certification practice exam page shows enterprise plus security now worth 30 to 40 percent of the paper.
Six months later the governance pair followed. GH-100's July rewrite added the Data Residency deployment scenario, Azure private networking for runners, third-party vault integration and metered-product cost optimization; the weighting on the GitHub Administration certification practice test page shows security, not user management, is now the biggest admin domain. GH-500's rewrite is the most visible of all because the domain titles themselves renamed: Secret Protection formerly secret scanning, supply chain security formerly Dependabot and Dependency Review, Code Security formerly code scanning with CodeQL. The GitHub Advanced Security certification practice test page includes the ten-second stale-prep test built on those renames.
GH-600 Developing in Agentic AI Systems is the family's first genuinely new exam in years: 120 minutes instead of the usual 100, no official practice assessment yet, and still in beta as of July 2026, meaning results are not scored immediately. Its outline is the densest we have measured on any vendor exam, with agent appearing 71 times in roughly 850 words. The full analysis is on the GitHub agentic AI certification practice test page, and the choice between the two AI exams is covered in GH-600 vs GH-300.
The wording is identical across the rewritten exams, and it is worth quoting because it defines how deep the changes go: the exam \"has changed significantly (e.g., new objectives were added, some were removed, existing objectives may have moved to different functional groups, and all were reworded).\" That last clause matters most for prep: even objectives that survived were reworded, which is why keyword-matching an old course syllabus against the new outline undercounts the drift. The stamps to look for on each study guide are \"Skills measured as of January 2026\" on GH-900 and GH-200, and \"Skills measured as of July 2026\" on GH-100 and GH-500.
Yes. A change to the exam outline changes what new candidates are tested on; it does not revoke credentials already earned. Check your certification's own renewal terms for when you personally need to act. The practical effect of a rewrite lands on your knowledge, not your badge: the platform the exam now describes has moved, and the gap is yours to close on your own schedule.
Measured by visible surface, GH-500: three of its six domain titles renamed, and prioritization logic changed with EPSS. Measured by audience impact, GH-900, because a fundamentals exam adding Copilot agents and Agent Mode means even non-developers studying for their first cert are now tested on AI tooling. GH-600 is technically the biggest change of all, being an entirely new exam.
Two forces explain the sweep. First, the product moved: agents, MCP, immutable actions, Push Protection and EPSS-driven triage all shipped or matured in the last two years, and exams that ignored them were certifying people on an old platform. Second, GitHub's certs live on Microsoft's exam infrastructure, which follows a cadence of dated skills-measured documents and published change logs. That cadence is a gift to candidates who use it: the change log tells you exactly when the ground moved, and the outline tells you exactly what is on the current paper.
Date-check everything. Before trusting any course or question bank, find its publication date and compare it against the change log of your exam. For GH-100 and GH-500 the cutoff is brutal: material from as recently as May 2026 predates the current outline.
Prefer generated questions over static banks in a rewrite year. A static bank ages the day the outline changes. Building questions from the current study guide itself, with a certification exam generator, means your drills inherit the new objectives automatically because the source document is the new blueprint.
For teams, treat 2026 as a recertification trigger. If your platform or security engineers certified on the old outlines, their credential still stands, but their knowledge of the tested material now trails the product. Teams that run structured internal training have an edge here; rolling the new study guides into a corporate learning platform and re-drilling the changed domains costs a few hours per engineer, far less than discovering the gap during an incident.
And if you are choosing your first GitHub cert, the rewrites are good news: the material you learn now matches the platform as it actually ships. Where to start, in what order, is mapped in our GitHub certification path guide.
Z tej samej rodziny narzędzi