Meagan AngelucciCDMP · SHRM-CP M.S. Industrial-Organizational Psychology

The free work

Is your AI use defensible?

A private check against CAN-ASC-6.2, Canada's voluntary standard for accessible AI. Nothing you type leaves your computer.

FreeYour answers are never saved or sent, and there is no sign-up. Visits are counted anonymously. If you allow analytics, it also counts only whether you finished and whether you copied, never what you chose. The privacy notice

This is not an audit and it is not a compliance check. CAN-ASC-6.2 is a voluntary Canadian standard, and it asks these things of organizations. What follows is one practitioner’s practice, read against the parts that land where a decision gets made.

Your own practice

If you cannot tell from the file, the answer is no. A file that cannot tell you is the finding.

  1. “Organizations shall inform people about their intended or actual use of AI systems that make decisions about, or otherwise impact, people with disabilities.” Clause 12.10 carries the same duty.

    1The person knew. The person whose file this is was told AI was used in their matter.

  2. Mechanisms “shall enable the people identified in 11.4.3 a) to request an equivalently full-featured and timely alternative decision-making process that is, at the individual’s choice, either performed: without the use of AI; or made using AI with direct human oversight and verification of the decision.”

    2They could have said no. They could have asked for the same decision without AI, or with a human checking it — and been told so.

  3. “People shall be given information in accessible formats about ways to: correct; contest; change; or reverse an AI-assisted decision or action that impacts them.” Clause 12.12 covers feedback, incidents and complaints.

    3They knew how to push back. They were given information, in a format they can use, on how to correct, contest, change or reverse the decision.

  4. “Organizations shall ensure there is a traceable chain of human responsibility that makes it clear who is accountable for the accessibility and equity of decisions made by an AI system.” And 11.2.6: organizations “shall not subject people with disabilities to AI-assisted decisions without their informed consent, understanding and access to equivalent alternatives.”

    4A human decided. Eligibility, credibility, fitness for work, or whether to accommodate was decided by a person, not produced by a tool and adopted.

  5. The same clause, read forward: traceability is not satisfied by a decision that is merely correct.

    5The file says who. Someone reading this file in a year could tell which parts a machine produced and which a human decided.

  6. Not a clause — the standard says nothing about verifying a summary. In a peer-reviewed clinical study that counted both against the same material, summaries dropped 3.45% of source sentences and added 1.47% of generated sentences: two rates on different counts, so read them side by side. Asgari et al., npj Digital Medicine, 2025, on mock primary-care consultations; its authors were employees of a clinical-scribe vendor when it was written.

    6You read the source. You read the underlying medical or functional-abilities document, not only the summary of it.

  7. “Organizations shall refrain from using AI tools for: biometric categorization; emotion analysis; or predictive policing of people with disabilities.” This covers the first half of the row.

    Effort, engagement and credibility are practice, not the clause.

    7Nothing inferred about the person. You did not use a tool to read emotional state from behaviour — and you did not use one to judge effort, engagement, consistency of presentation, or credibility.

  8. “Organizations shall ensure that disability data is used to create equitable AI systems in a safe and secure manner that does not result in harms for the individuals who provide their data.” And 12.14.2: “Organizations shall ensure that re-identification is not possible through strong anonymization techniques.”

    8Nothing identifying went somewhere it should not. No name, no identifying detail and no third party’s information went into a tool your organization has not approved for it.

8 checks

Your organization’s, and you may not know

“Don’t know” is a true answer here, and it is counted on its own, never as a failure.

  1. “The intention to use an AI system shall be publicly disclosed in accessible formats as part of the organization’s accessibility plan and distributed to national disability organizations and interested parties.” The clause names the document, so this one can be checked. And 10.2.1: transparency information “shall be published before deployment and kept up-to-date”.

    9Your employer’s accessibility plan says so. Your organization’s published accessibility plan discloses that it uses AI in this kind of decision.

  2. “The organization shall address accessibility and equity-related feedback, incidents, and complaints, including providing details about redress, challenge, and appeals functions for people with disabilities.” And 10.2.2: “Mechanisms to provide feedback … shall be accessible to people with disabilities.”

    10There is somewhere to complain. The person could raise a problem about the AI use itself, not just about the outcome.

  3. Organizations that deploy AI systems shall provide alternative options that offer “equivalent availability, reasonable timeliness, cost and convenience to the person with a disability”, including final determination “by a human with expertise in the needs of people with disabilities”. “The organization shall retain individuals that have the necessary expertise to make equitable human decisions regarding people with disabilities when AI systems are deployed to replace decisions previously made by humans.”

    11A real alternative exists. If someone asked for a decision without AI, your organization could actually provide one at comparable speed and cost.

  4. “All personnel responsible for any aspects of the AI lifecycle shall receive training in accessible and equitable AI, delivered in accessible formats.”

    12You were trained for this. You have had training on using AI in decisions that affect people with disabilities.

4 checks

It only flags problems. A line that held stays plain.

Checked againstCAN-ASC-6.2:2025, Accessible and equitable artificial intelligence systems. Accessibility Standards Canada, December 2025. Free and bilingual.

Left out, on purposeEverything about building, training, procuring, customizing or terminating a system — clauses 10.1.x, 11.1.x, 12.4 through 12.8, 12.13. Roughly fifty requirements. They are real and they bind organizations; they are not answerable by the person using the tool. Disaggregated performance metrics, bias audits, public registries of harm (11.1.3, 12.8.2). They cannot be a self-check. Anything requiring a conformance verdict. The standard is “a reference document for voluntary use”, and Accessibility Standards Canada “has no power, nor does it undertake, to enforce conformance.”

What it never saysThat you are compliant, non-compliant, certified, assessed or at risk. Whether any law or regulation makes this standard mandatory for your organization is a question for counsel.

How the wording was checkedEvery line in quotation marks was read against the standard’s published full text on 27 September 2026 and matches it word for word. What sits outside the quotation marks is a reading of the clause, not the clause.