AI workflow assurance
Evidence from scoped checks of an existing workflow, with findings, tested limits, and next steps.
Relevant to a scoped workflow assessmentA guide to the language
AI assurance and discovery glossary
Plain-language terms for understanding workflow assessments, related business needs, and AI discovery. Service links indicate the nearest topic, not a promise that every listed activity is included.
These terms describe overlapping goals and different kinds of work. A term appearing here does not mean every activity is included in an assessment. The scope determines what is tested and what evidence you receive.
112 terms shown
What the categories mean
These concepts answer different questions. A service link identifies the nearest topic; its fit label defines the relationship.
AI assurance
Evidence supporting a specific claim about an AI system, communicated with its limitations.
Related practice; agree the scopeAI governance
Rules, responsibilities, and decision processes for how AI is developed and used.
Related practice; agree the scopeAI risk management
Identifying possible harms or failures, deciding which matter, and managing the response.
Related practice; agree the scopeAI evaluation
Measuring system behavior against stated criteria using selected examples or methods.
Related practice; agree the scopeAI observability
Records of execution, such as logs and traces; these help investigation but do not establish correctness.
Related practice; agree the scopeResponsible AI
Principles and practices for considering how AI affects people throughout its use.
Related practice; agree the scopeCompliance
Meeting applicable obligations. A workflow assessment alone does not establish legal compliance.
Background terminology; not a Guth certification or legal serviceCertification
A formal determination under a particular scheme. Guth's workflow assessment does not award certification.
Background terminology; not a Guth certification or legal serviceAssessing an existing workflow
These phrases lead to a bounded assessment of agreed sources, actions, and test scenarios.
AI assurance services
Scoped assessment work that produces findings and supporting evidence about an existing AI workflow.
Relevant to a scoped workflow assessmentAI workflow assessment
Reviewing one workflow's sources, permissions, and behavior against agreed criteria.
Relevant to a scoped workflow assessmentAI agent audit
A technical assessment of an agent's behavior and permitted actions; here, audit means workflow review.
Relevant to a scoped workflow assessmentAI agent assessment
Another name for a scoped technical review of an agent and the workflow around it.
Relevant to a scoped workflow assessmentAI risk assessment services
Assessing likely workflow failures and consequences to prioritize checks and controls.
Relevant to a scoped workflow assessmentagentic AI assurance
Assessment of systems that select steps or use tools, including the limits on their actions.
Relevant to a scoped workflow assessmentRules and responsibility
Workflow-level governance review and organization-wide implementation have different scopes.
AI governance consulting
Advice on ownership, rules, and review processes; an organization-wide engagement needs its own scope.
Separately scoped; availability must be confirmedAI governance assessment
Reviewing whether stated responsibilities, rules, and oversight arrangements are clear and evidenced.
Related practice; agree the scopeAI governance implementation
Putting governance processes and controls into operation; implementation requires a separate agreed scope.
Separately scoped; availability must be confirmedAI governance framework
An organized set of roles, policies, controls, and review processes for AI use.
Related practice; agree the scopegenerative AI governance
Governance applied to systems that generate content, including source use, review, and publishing responsibilities.
Related practice; agree the scopeagent governance
Rules about an agent's tools, permissions, decision limits, and accountable human owners.
Related practice; agree the scopeagentic governance
A market phrase for governance of systems that plan or act through tools; closely related to agent governance.
Related practice; agree the scopeOwners, approvals, and records
These terms describe controls or evidence that can be examined within an agreed workflow.
AI oversight
How responsible people review AI behavior and intervene when needed.
Relevant to a scoped workflow assessmentAI accountability
Who is answerable for decisions about the workflow and for responding when it fails.
Relevant to a scoped workflow assessmenthuman oversight of AI
The human review and intervention points built into an AI workflow.
Relevant to a scoped workflow assessmentAI controls assessment
Checking whether specified safeguards exist and behave as expected in selected cases.
Relevant to a scoped workflow assessmentAI approval workflows
The steps requiring a person's authorization before an AI-generated action proceeds.
Relevant to a scoped workflow assessmentAI autonomy controls
Limits on which decisions and actions an agent can take without further approval.
Relevant to a scoped workflow assessmentAI audit trail
A reviewable record of inputs, actions, approvals, and outcomes; its completeness must also be checked.
Relevant to a scoped workflow assessmentTesting behavior
A test result applies to the cases, criteria, versions, and conditions recorded with it.
LLM evaluation
Measuring language-model behavior against chosen criteria; model-level results may differ from whole-workflow results.
Relevant to a scoped workflow assessmentLLM testing
Running selected cases to observe language-model outputs and failures; one activity within a broader evaluation.
Relevant to a scoped workflow assessmentAI agent evaluation
Testing an agent's decisions, tool use, and outcomes against stated expectations.
Relevant to a scoped workflow assessmentchatbot testing
Checking a conversational workflow's responses, source use, and handoffs with representative questions.
Relevant to a scoped workflow assessmentAI validation
Checking whether evidence supports suitability for an intended use and its acceptance criteria.
Relevant to a scoped workflow assessmentAI reliability testing
Checking consistency and failure patterns across repeated or varied conditions.
Relevant to a scoped workflow assessmentagent regression testing
Rerunning established cases after a change to detect behavior that has worsened.
Relevant to a scoped workflow assessmentAI requalification
Rechecking whether earlier acceptance still holds after relevant changes; criteria and evidence must be stated.
Relevant to a scoped workflow assessmentSources and retrieval
Finding a document, quoting it, and correctly applying it are separate checks.
RAG evaluation
Testing retrieval-augmented generation: which material is retrieved and how the answer uses it.
Relevant to a scoped workflow assessmentretrieval quality
How well retrieved material matches the question, including relevance, coverage, and freshness.
Relevant to a scoped workflow assessmentAI grounding
Connecting an answer to appropriate evidence or context; the connection still needs checking.
Relevant to a scoped workflow assessmentcitation verification
Checking that a cited source exists and supports the associated claim.
Relevant to a scoped workflow assessmentsource provenance
The origin and history of a source, including version and transformations when available.
Relevant to a scoped workflow assessmentAI knowledge accuracy
Whether the information used or produced is correct for the question and context being tested.
Relevant to a scoped workflow assessmentstale knowledge
Outdated information that may produce an answer inconsistent with current policy or facts.
Relevant to a scoped workflow assessmentAI data quality
The fitness of data for its intended use, including correctness, completeness, and relevant context.
Relevant to a scoped workflow assessmentPermissions and security vocabulary
These are distinct security topics. Specialist work needs explicit scope, authorization, and a competence check.
AI red teaming
Adversarial testing designed to uncover failures; a scoped exercise cannot demonstrate universal security.
Separately scoped; availability must be confirmedprompt injection testing
Testing whether untrusted instructions can redirect a model or its tools beyond intended behavior.
Separately scoped; availability must be confirmedagent permission testing
Checking whether an agent can access only the resources and actions it has been authorized to use.
Separately scoped; availability must be confirmedtool-use safety
Examining hazards when an AI system invokes tools, including authorization and consequences.
Separately scoped; availability must be confirmedexcessive agency
A risk arising when a system has more functionality, permission, or autonomy than its task needs.
Background terminology; not a Guth certification or legal serviceAI data leakage testing
Testing agreed scenarios in which information might be disclosed to an unauthorized destination.
Separately scoped; availability must be confirmedChange and ongoing operation
A one-time assessment does not include an ongoing monitoring or incident-response service.
AI monitoring
Ongoing observation of selected signals and failures, requiring separately agreed coverage and response duties.
Separately scoped; availability must be confirmedmodel drift
A change affecting model performance as data, conditions, or the model itself change over time.
Related practice; agree the scopebehavioral drift
Observed changes in system behavior, potentially caused by tools, prompts, policies, or sources as well as models.
Related practice; agree the scopecontinuous AI assurance
Recurring evidence collection and review; cadence, coverage, and ownership require separate agreement.
Separately scoped; availability must be confirmedAI change management
Recording, reviewing, and testing changes before and after they affect an AI workflow.
Related practice; agree the scopeAI lifecycle governance
Responsibility and review processes spanning introduction, operation, change, and retirement of AI.
Related practice; agree the scopeAI incident management
Processes for reporting, investigating, containing, and learning from AI failures; response services need separate scope.
Separately scoped; availability must be confirmedPrinciples and affected people
These qualities are related, but testing one does not establish all of them.
responsible AI consulting
Advice on translating responsible-AI principles into practices; broader programs require separate qualification.
Separately scoped; availability must be confirmedtrustworthy AI
An umbrella description of desired qualities that must be supported by specific evidence in context.
Background terminology; not a Guth certification or legal serviceethical AI
Considering values and effects on people when deciding how AI should be developed or used.
Background terminology; not a Guth certification or legal serviceAI transparency
Making relevant information about an AI system and its use understandable and available.
Background terminology; not a Guth certification or legal serviceexplainable AI
Methods or information intended to help people understand an output or behavior; explanations also need scrutiny.
Background terminology; not a Guth certification or legal serviceAI fairness
A context-dependent concern about equitable treatment and outcomes, requiring explicit criteria and affected-group consideration.
Background terminology; not a Guth certification or legal serviceAI bias assessment
Examining specified forms of bias with suitable evidence and methods; it does not prove universal fairness.
Separately scoped; availability must be confirmedalgorithmic accountability
Responsibility and mechanisms for examining, challenging, and addressing decisions influenced by algorithms.
Background terminology; not a Guth certification or legal serviceBusiness and supplier review
These terms help scope a conversation about suppliers or internal controls. Organization-wide programs exceed a single workflow assessment.
AI vendor assessment
Reviewing a supplier's documented capabilities, limitations, terms, and evidence for an intended use.
Separately scoped; availability must be confirmedthird-party AI risk
Risks introduced through external AI suppliers, models, data, or connected services.
Related practice; agree the scopeAI due diligence
A structured investigation of relevant AI claims and dependencies before a business decision.
Separately scoped; availability must be confirmedAI inventory
A maintained list of AI systems and uses, with owners and relevant dependencies.
Related practice; agree the scopemodel inventory
A record of models and versions in use; narrower than an inventory of complete AI workflows.
Related practice; agree the scopeAI risk register
A record of identified risks, owners, treatments, and review status.
Related practice; agree the scopeAI policy review
Checking the clarity and implementation of internal AI rules; legal review is a separate professional question.
Related practice; agree the scopemodel risk management
Managing risks arising from model use, including assumptions, limitations, validation, and change.
Separately scoped; availability must be confirmedAI GRC
Governance, risk, and compliance activities concerning AI; a broad program label, not a single assessment.
Background terminology; not a Guth certification or legal serviceStandards and legal vocabulary
Educational definitions for scoping discussions. Listing a standard does not claim certification, conformity, or legal advice.
NIST AI RMF assessment
Reviewing practices against selected parts of NIST's voluntary AI Risk Management Framework; this is not government approval.
Background terminology; not a Guth certification or legal serviceISO/IEC 42001 readiness
Preparation against requirements of the AI management-system standard; readiness work is distinct from certification.
Background terminology; not a Guth certification or legal serviceISO/IEC 23894
An international standard providing guidance on AI risk management.
Background terminology; not a Guth certification or legal serviceAI impact assessment
Examining potential effects of an AI system on people and its surrounding context.
Background terminology; not a Guth certification or legal servicealgorithmic impact assessment
Examining effects of algorithmic decision-making; methods and obligations depend on the context.
Background terminology; not a Guth certification or legal serviceEU AI Act readiness
Preparation for potentially applicable EU AI Act duties; determining obligations requires qualified legal analysis.
Background terminology; not a Guth certification or legal serviceAI conformity assessment
Evaluating whether specified requirements are met under a relevant process; this is not synonymous with a general workflow review.
Background terminology; not a Guth certification or legal serviceOther labels you may encounter
These phrases can overlap with assessment work, but they do not establish a common scope or assurance level.
AI trust management
A broad label for practices intended to support justified confidence in AI; specific evidence and controls still matter.
Related practice; agree the scopeAI TRiSM
Gartner's label for AI trust, risk, and security management, spanning several technical capabilities and disciplines.
Background terminology; not a Guth certification or legal serviceAI quality assurance
Planned checks and processes for maintaining specified quality criteria in an AI workflow.
Related practice; agree the scopeAI safety testing
Testing selected possibilities of harm under agreed conditions; specialist or high-stakes work needs separate scope.
Separately scoped; availability must be confirmedAI assurance case
A structured argument connecting a defined claim to supporting evidence, assumptions, and limitations.
Related practice; agree the scopeAI control framework
An organized set of controls tied to risks, responsibilities, and evidence requirements.
Related practice; agree the scopeAI governance maturity assessment
Reviewing governance practices against a stated maturity model; the model and evidence determine what the result means.
Separately scoped; availability must be confirmedQuestions businesses ask
Practical starting points for choosing what to assess.
how to check AI answers
Compare claims with appropriate current sources and record unsupported statements, errors, and uncertainty.
Relevant to a scoped workflow assessmentcan my chatbot use old policies
It may retrieve obsolete material; test conflicting versions and check how the current policy is selected.
Relevant to a scoped workflow assessmenthow to test an AI agent before launch
Define permitted actions and acceptance criteria, then run representative and failure cases with recorded results.
Relevant to a scoped workflow assessmentstop an AI agent sending without approval
Review whether sending is blocked by enforceable authorization until the required person approves.
Relevant to a scoped workflow assessmenthow to audit an AI workflow
Inventory sources and actions, agree test criteria, and record observations, limitations, and next steps.
Relevant to a scoped workflow assessmentevidence for AI decisions
Identify the source, relevant inputs, action history, and approval record needed to review a particular outcome.
Relevant to a scoped workflow assessmentWorking with implementers
These review topics can support a customer handoff. Scope and reviewer relationships should be clear.
AI agency quality assurance
Review of agreed quality checks around an agency-built workflow; review independence must be disclosed.
Relevant to a scoped workflow assessmentAI implementation review
Examining how an implemented workflow matches documented requirements and permission boundaries.
Relevant to a scoped workflow assessmentagent handoff checklist
A record of owners, configuration, limits, test evidence, and operating instructions needed for a handoff.
Relevant to a scoped workflow assessmentcustomer AI acceptance testing
Testing against criteria agreed with the customer before they accept the workflow.
Relevant to a scoped workflow assessmentAI deployment review
Examining release evidence, configuration, access, and rollback arrangements; it does not certify production readiness.
Relevant to a scoped workflow assessmentSearch and AI discovery
These marketing labels overlap. Crawlability, indexing, citations, visits, and inquiries are different outcomes.
SEO
Search engine optimization: improving a site's accessibility, relevance, and presentation for search discovery.
Related practice; agree the scopeAEO
Answer engine optimization: a marketing label for making content useful and understandable in answer-based experiences.
Related practice; agree the scopeGEO
Generative engine optimization: a marketing label for improving how content can be found and used in generated answers.
Related practice; agree the scopeAI search optimization
Work on content and technical discoverability for AI-supported search; there is no universal submission channel.
Related practice; agree the scopeLLM visibility
A broad label for whether information about a business appears in language-model experiences; measurement depends on the experience.
Related practice; agree the scopecrawling
Fetching pages or resources. A crawler visit does not establish indexing or a recommendation.
Related practice; agree the scopeindexing
Processing content into a search system's index; crawl access alone does not guarantee inclusion.
Related practice; agree the scoperetrieval
Selecting information to use for a query, from an index, direct fetch, or connected source.
Related practice; agree the scopeanswer selection
An answering system's choice of material and claims to include; discovery alone does not determine that choice.
Related practice; agree the scopecitation
A source reference in an answer; its presence is distinct from a visit, endorsement, or sale.
Related practice; agree the scopereferral visit
A visit arriving through another site or experience; attribution may be incomplete.
Related practice; agree the scopeinquiry
A person's request for information or help; it is a separate event from a citation or visit.
Related practice; agree the scope