What it is for
Many governments now require AI literacy training. Far fewer mandates change what the workforce can actually do. The guide argues that the difference lies mainly in how a mandate is designed, and it is written for officials who already have the authority to issue a requirement and are asking how to do it well. A draft version 1.0 exists; it is informative guidance, not a requirement of the Standard.
Five ways mandates fail
- Undefined: requires ‘AI literacy’ without saying what it means, who is in scope or what level is needed.
- Unfunded: sets requirements without resources, so costs fall entirely on the organizations that must comply.
- Unverified: does not say how compliance will be evidenced or checked.
- Unmaintained: tied to a particular course rather than a versioned competency standard, so it goes out of date.
- Shallow: requires ethics or safety training without the data literacy that makes it meaningful.
The guide’s central point is that a narrower, well-enforced mandate achieves more than a broad, aspirational one.
What the guide covers
- The policy instrument: directives, regulatory standards or strategic commitments, and how to reference ALiF in each (by version, with an update clause, or by level).
- The population in scope: defined by what people do with AI outputs, not by job title, with minimum levels that rise with direct AI use and responsibility.
- Compliance tiers: a phased structure over about three years, moving from baseline and governance commitment, to demonstrated competence in higher-risk groups, to sustained conformance.
- Realistic timelines: allowing for assessment, program delivery and infrastructure build, with guidance on extensions.
- Reporting and verification: aggregate returns, visible spot-checks and an annual sector-level report.
- Early warning signs: and the risk that powerful organizations soften requirements, countered by transparency.
- A worked example: from an academic health system, which the guide itself says is illustrative rather than fully operational.
Why it is being revised
The draft relies on self-assessment results as evidence that people have reached a level. Under the proposed ALiF 2.0 conformance requirements, self-assessment results could not be used in decisions about individuals, and self-report instruments would carry a disclosure duty. No validated ALiF instrument that goes beyond self-report exists yet, so version 2.0 will need to say plainly what mandates can and cannot evidence at present. It will also encourage mandating authorities to require a Learning Conditions Statement from organizations in scope, and reflect Data Literacy First.