By Lucia Elliott, ACA Head of Marketing & Sponsorship, and AWARD Lab GenAI Sprint Lead.
We’ve spent a lot of time talking about taste in AI, and for good reason. How human creative direction brings discernment, judgment, and aesthetic nuance to synthetic output is what separates mediocre execution from great work. But while taste has been central to much of the discussion, arguably the more urgent commercial challenge is governance.
The most pressing existential friction facing agencies today isn’t just whether an AI output delivers quality—it’s whether we can legally sell it, guarantee client ownership, and protect the agency from liability along the way.
When AWARD recently asked 25 Australian CCOs and ECDs about their primary compliance concerns, the anxiety wasn’t localised; it was an all-fronts risk minefield:
- 60% cited IP and copyright ownership of AI-generated assets
- 56% worried about brand integrity and visual consistency
- 52% cited direct legal liability for AI-generated content
- 52% flagged data privacy and confidentiality
- 52% feared reputational risk and public backlash
In parallel, we’ve been watching a global regulatory net tighten in real time. In the US, courts continue to uphold that purely autonomous AI outputs cannot hold copyright without proven, substantial human authorship. In Europe, the EU AI Act’s Article 50 transparency obligations enforce strict machine-readable marking and watermarking for synthetic content. Closer to home, the Australian Government explicitly ruled out introducing a Text and Data Mining (TDM) copyright exception, meaning platform scraping without consent remains a legal quagmire. No wonder we’re a bit spooked.
Yet despite these confronting statistics, some agency AI and client strategies consist of little more than a PDF policy document that few teams ever read during a brief.
On the flip side, hackathons and agency AI sprints have historically ended at a conceptual slide deck, where creatives have been free to experiment and fail safely by generating cool prototypes showcased on a slide and presented in a sandbox. But when you move AI out of a controlled environment and into live commercial deployment, the rules change entirely.
That was the catalyst for the inaugural AWARD Lab GenAI Sprint, which was conceived as an industry test case to examine what happens when AI meets real-world delivery constraints. Teams worked with Leonardo.Ai on a live behavioural brief for client AUSVEG (+ONESERVE), with winning work progressing directly to national Digital Out-of-Home (DOOH) deployment via the Outdoor Media Association’s member network.
Because the end product was destined for public screens, by necessity and professional rigour, AWARD built a ground-up, governance-first operational framework to make AI-enabled creative work not only human-directed, but commercially viable and legally defensible:
- Weighted strategic dominance: Structurally prioritising core human strategic intent and conceptual ingenuity over raw algorithmic execution, ensuring prompt luck could never displace genuine creative mastery.
- Traceable chain-of-custody logging: Implementing full-screen, timestamped process capture to establish verifiable proof of human directional authorship and prompt iteration.
- Multi-layer asset provenance: Enforcing mandatory verification protocols auditing visual origins across client libraries, authenticated original source files, and pre-cleared third-party assets.
- Real-time craft & channel integration: Embedding expert production specialists directly into the workflow to ensure aesthetic realism, technical channel constraints, and legal boundaries were managed from inception.
- Quantified ethics & bias guardrails: Integrating explicit representation, fairness, and algorithmic bias checks directly into formal review criteria.
Crucially, what we tested through AWARD Lab did more than just solve an immediate operational workflow—it started to make the case for how we judged AI-enabled work and informed the way we constructed our judging criteria as well.
Admittedly, enforcing this level of governance isn’t frictionless. Tracking process logs, auditing source files, and verifying chain-of-custody requires real operational effort and administrative discipline from both production and account teams. It adds steps to a workflow that AI was supposed to speed up.
Yet the results of the AWARD Lab sprint yielded a fascinating counter-narrative to common industry assumptions. While the C-suite may be gnashing its teeth over liability, mid-weight creatives aren’t resisting structure—they’re starving for it. In our post-sprint participant survey, teams valued creative leadership and clear process guardrails far higher than technical tool instruction.
As AWARD Council member and Leo Australia ECD Hilary Badger noted in her opinion piece on the sprint, clear process guardrails and mandatory logging didn’t stifle creativity. They gave teams explicit permission to fail, granting them the room to experiment without fear.
Governance isn’t something that comes after creativity. Governance is the creative framework. By standardising process logging, asset provenance, and platform liability carve-outs, the AWARD Lab framework proved that GenAI can be transformed from an unquantifiable corporate risk into a scalable, production-ready creative engine.
In the end, solving the governance problem doesn’t diminish the role of taste and human judgment. By embedding clear boundaries early, we give great taste the safe, scalable, and legally defensible foundation it needs to actually work.
It’s time to move past taste. The future of creative tech belongs to those who build the guardrails to drive fast, and safely.
Footnotes & Sources:
- Thaler v. Perlmutter, No. 23-5233 (D.C. Cir.); US Copyright Office Guidance on Works Containing Material Generated by Artificial Intelligence.
- European Union AI Act, Regulation (EU) 2024/1689, Article 50 (Transparency obligations for providers and deployers of certain AI systems).
- Attorney-General’s Department (Australia), Government Response to AI and Copyright Consultation, ruling out statutory Text and Data Mining (TDM) exceptions under the Copyright Act 1968 (Cth).
