InfoQ Launches Certification Cohorts on AI Security and Coding Agent Verification

InfoQ is running two online certification cohorts aimed at two of the thornier problems teams face when AI moves past prototypes: security and privacy choices in production systems, and how to verify what coding agents actually change. Both programs are being highlighted by InfoQ contributor Artenisa Chatziou. The pairing reflects a shift in what AI practitioners are being asked to prove — not just that a model works, but that it can be trusted inside systems that already carry real data and real users.

The security cohort centers on decisions that surface only after deployment: how data flows through model pipelines, where privacy boundaries sit, and which controls hold up under production load. Those questions have grown more urgent as companies push generative features into customer-facing products, where a single misconfigured retrieval store or over-permissive tool call can expose sensitive records. The curriculum frames security as a set of engineering trade-offs rather than a compliance checkbox, which matches how most teams actually encounter the problem.

The second cohort tackles verification for coding agents that operate inside existing repositories. When an agent edits a codebase it did not write, the hard part is not generating a diff — it is confirming the change does what was asked without breaking adjacent behavior. That means tests, review gates, and tooling that can trace an agent's edits back to a stated intent. The topic has moved quickly from novelty to daily engineering concern as more teams wire agents into CI pipelines and internal tooling.

InfoQ's decision to package both subjects as certification cohorts says something about the maturity curve. Training that once focused on model selection and prompt design now has to cover operational failure modes, because that is where budgets and reputations are being lost. Certification also gives employers a way to signal competence in areas that lack settled standards, though the value of any credential depends on whether the skills transfer to real incidents.

What to watch is whether these cohorts produce durable practices or just vocabulary. The verification side especially needs shared benchmarks — what counts as a safe agent edit, and who signs off when an automated change ships. If the industry settles on answers, programs like these will look prescient. If not, they will be another round of training that outpaces the tooling it describes.