You cannot govern what you cannot see. This topic covers the tooling that records what an AI system actually did.
Tracing, evaluation, logging, and audit trails for AI systems in production.
24 stories
AI Governance & Ethics
A fluent explanation is useless if it cannot be traced to the rule that produced a denial. Reason codes make chatbot language reviewable before it reaches a customer.
Irene Vasko · October 4, 2026 · 8 min
AI Governance & Ethics
One privacy request can touch four separate data layers. Each needs its own remedy, owner, and evidence, while model weights may require testing, unlearning, or retraining.
Irene Vasko · 8 min read
AI Governance & Ethics
A permit-review prompt, its attachments, model output, and staff edits can carry different retention and disclosure duties. Agencies need a retrieval workflow before the first request arrives.
Irene Vasko · 8 min read
AI Industry & Models
A cheaper model can handle routine requests, but one bad routing decision may create retries and manual review. Logging the decision makes the real savings measurable.
Tobias Lund · 7 min read
AI Governance & Ethics
A defensible denial record connects the customer’s inputs, model result, business rule, human action and notice. New Jersey law does not reduce that chain to one AI log.
Irene Vasko · 8 min read
AI Industry & Models
Real-time voice models can remove transcription from the application path. For support calls that need moderation and audit logs, developers may still need to generate text beside the conversation.
Tobias Lund · 8 min read
AI Governance & Ethics
A small company does not need a binder of AI policies. It needs retrievable records showing who approved the system, how it was tested, what counts as an incident, and what changed.
Irene Vasko · 8 min read
AI Governance & Ethics
Colorado’s AI law turns ordinary system documentation into compliance evidence. Start with one record per decision workflow, including its inputs, reviewers and vendor dependencies.
Irene Vasko · 8 min read
AI Governance & Ethics
A policy statement cannot prove that an AI control operated. Auditable governance ties each model release to tests, ownership, approval, deployment records, and a defined stop condition.
Irene Vasko · 8 min read
AI Governance & Ethics
A model card cannot explain why a customer was denied a credit limit increase. This decision-level template connects affected people, harms, controls, evidence, and appeals.
Irene Vasko · 8 min read
AI Governance & Ethics
A score cannot show whether bad source data, a model transformation, a policy threshold, or a human reviewer caused a denial. Preserve a decision-time record that can separate them.
Irene Vasko · 7 min read
AI Governance & Ethics
A polished denial letter can be detached from the model and rules that rejected an applicant. The lender needs a decision-time trace, not a story generated from the result.
Irene Vasko · 8 min read
AI Governance & Ethics
A generic denial notice can conceal a gap between a model’s inputs and its stated reasons. This workflow tests reason codes against the decision record before notices go out.
Irene Vasko · 8 min read
AI Governance & Ethics
A ranking score, generated summary and recruiter click can form one automated decision path. Employers need records showing each step, its owner and whether a human could reverse it.
Irene Vasko · 8 min read
AI Governance & Ethics
New scoring inputs, thresholds or model weights can make an annual bias audit poor evidence for the tool now screening applicants. The deployment log should show whether the audited and operating systems still match.
Irene Vasko · 8 min read
AI Governance & Ethics
A language model can make decision codes readable, but fluency is not evidence. The safe workflow binds every sentence to factors the decision system recorded as principal reasons.
Irene Vasko · 8 min read
AI Governance & Ethics
Hosted models, retrieval data, and safety controls can change outputs while application code stays fixed. Treat each dependency as a versioned production component.
Irene Vasko · 8 min read
AI Industry & Models
A retirement date tells you when an endpoint closes, not whether its replacement behaves the same. Shadow traffic exposes changes in refusals, tool calls, latency, and length before cutover.
Tobias Lund · 8 min read
Agentic AI & Orchestration
An agent can change an inbox, lose its working context, then repeat the same actions. Checkpoint design determines whether it resumes cleanly or leaves duplicate drafts and missing work.
Mara Quintero · 7 min read
Agentic AI & Orchestration
On a recurring compliance form, Playwright stopped when a field changed. The browser agent kept going, which reduced repair work but made a wrong document upload harder to catch.
Mara Quintero · 7 min read
AI Governance & Ethics
A chat transcript is not an audit trail. Before an agent submits a payment, its log must preserve the evidence, authorization, and exact action that crossed the boundary.
Irene Vasko · 8 min read
AI Governance & Ethics
A deletion receipt from the application is only the first step. Real evidence must follow the prompt through provider logs, backups, vector databases, and tracing systems.
Irene Vasko · 8 min read
AI Industry & Models
Valid JSON is only the first gate. Production systems must also catch truncated responses, schema drift, unsupported fields, and values that look valid but trigger the wrong action.
Tobias Lund · 8 min read
AI Governance & Ethics
A defensible approval record shows the evidence presented, the reviewer’s changes and the rule behind an exception. A name and timestamp cannot prove meaningful oversight.
Irene Vasko · 8 min read
Other impacts
One story a day
One real story about AI each morning — no hype, no alarm, just company for the road.