Responsible AI, AI Governance & Ethics Explained | Marketing Stack 2025
In this session of Marketing Stack 2025, Abhinav Mittal, a published author, digital transformation leader, and advisor to CXOs with over 20 years of experience, discusses the critical intersection of AI adoption, ethics, and governance. Having delivered more than 45 digital transformation projects and unlocked over $100 million in cost savings, Mittal argues that while AI's creative potential is vast, its "ungoverned" use poses an existential threat to brands and marketing careers alike.
What is the difference between AI regulation and AI governance?
Abhinav distinguishes between these two concepts by framing regulation as "what you must do" and governance as "what you should do." Regulations, such as the EU AI Act or GDPR, are reactive, government-imposed laws with hard lines that carry legal consequences if crossed.
In contrast, governance is a proactive, internal set of policies and ethical guidelines that an organisation chooses to follow. Effective governance ensures that regulatory compliance becomes a natural outcome rather than a forced hurdle, ultimately building "trust," which Mittal identifies as the most valuable currency in the AI era.
What are the biggest risks of using ungoverned AI in marketing?
Beyond common concerns like bias, misinformation, and data privacy, Mittal identifies a more insidious risk: the loss of a marketing team’s "creative edge". He shares a case study of a large Fintech company that saw a 60% drop in acquisition costs after adopting a third-party AI platform.
However, when the tool went offline, the marketing team realised they had effectively "outsourced their brains" to the algorithm and could no longer explain their own creative strategy. This "hostage of AI" scenario highlights how over-reliance on ungoverned tools can hollow out human expertise.
Who is accountable when AI outputs or third-party tools fail?
Mittal is clear that while platform providers must offer safe tools, the ultimate accountability for AI usage rests with the brand and its marketers. He warns against being seduced by "flashy demos" and instead advocates for a rigorous vetting process.
Marketers should ask vendors difficult questions regarding their training data sources, the transparency of their "black box" algorithms, and their commitment to ethical standards. Furthermore, internal legal teams must be involved to address liability and data ownership issues before agreements are signed.
What core principles and frameworks should guide AI governance?
With over 170 AI governance frameworks in existence, Mittal recommends focusing on five core "Responsible AI" principles: fairness, transparency, accountability, privacy, and safety. Organisations seeking a roadmap should consider global standards such as ISO 42001, the NIST AI Risk Management Framework, or IEEE standards.
Notably, ISO 42001 is emerging as a universal benchmark, with major corporations like Microsoft now mandating it for their AI suppliers.
How can CMOs balance the speed of AI adoption with responsibility?
CMOs are urged to transition into the role of "Chief Ethics Officer" for the customer experience. This involves moving beyond technical implementation to establish an AI governance committee and documenting clear principles.
Key checkpoints include conducting risk assessments across all algorithms, institutionalising an incident management process, and ensuring AI literacy for all team members, from senior leaders to junior executives, to prevent accidental data breaches or breaches of confidentiality.
How can brands scale AI safely while maintaining human oversight?
While "human-in-the-loop" is a standard recommendation, Mittal notes it is often not scalable. He proposes a three-layer defence for marketing:
- Control Inputs: Fine-tune base models on your own curated brand corpus rather than using generic models for critical tasks.
- Use RAG: Implement Retrieval Augmented Generation (RAG) to force AI to pull information only from trusted, verified sources.
- Classifier Models: Pass AI outputs through a classifier model to automatically approve or reject content; only rejected content then requires mandatory human review, allowing the system to scale efficiently.
What are the implications of the EU AI Act and "Shadow AI"?
The EU AI Act is setting a global template with massive fines up to 35 million Euros or 7% of global revenue, and applies to any brand targeting EU consumers.
It requires strict documentation for high-risk use cases and mandatory labelling of AI-generated content. Simultaneously, Mittal warns of "Shadow AI", the unauthorised use of unapproved AI tools by employees.
To mitigate the risk of sensitive data leaking into these tools, CMOs must provide secure, white-labelled alternatives and maintain clear policies regarding unapproved software.
Ultimately, Mittal’s primary advice for the "perfect AI governance playbook" is to "innovate at the speed of trust, not at the speed of technology," as technology can be purchased, but trust must be earned.
