Agentic AI is promising speed, scale and efficiency. But at this year’s Cannes Lions, conversations kept coming back to the risks: Is it brand safe? Is data secure? Can we trust it?
In the second half of our roundtable discussion in partnership with Adform and Adsquare (read part one here), industry leaders explored how the adoption of agentic tools by agencies and brands comes with greater responsibility.
While there is a lot of excitement around automation, participants were clear-eyed about what happens when the human is removed from the loop.
Automation needs guardrails
For years, digital advertising has struggled with brand safety, whether due to ad placements next to harmful content or the monetisation of low-quality made-for-advertising websites. As agencies automate the media process at massive scale, getting the guardrails right matters more than ever.
Anna Campbell, Global MD at Carat, emphasised that automated execution must be accompanied by equally robust guardrails to protect everything from creative placement to brand equity.
“An AI system doesn’t choose what it accelerates; it accelerates everything, whether it is a placement or a mistake,” Campbell said.
She pointed out that while client leaders are eager to see cost savings and quick delivery, they also understand the potential impact of a media-buying error. This creates organisational tension, where leadership demands automation but remains highly risk-averse in its execution.
Luc Benyon, Marketing Director, Adsquare, pointed out the importance of ensuring AI is embedded across the organisation. “We’re integrating AI in our processes and our products. Making sure data is created, tagged and described in an AI-centric way will be a key skill going forward.”
Data security, privacy, and intellectual property
Beyond operational risks, the biggest questions around agentic AI were centred around all important data.
When agencies and brands are using agentic AI they are feeding it valuable information – from target segment profiles to campaign strategies and sensitive corporate data. That brings real questions around intellectual property exposure and privacy compliance. As a result, InfoSec reviews have become the critical checkpoint for agency innovation, with corporate security officers playing an important role in safeguarding proprietary data assets.
Solving these privacy concerns is paramount, said Katsur: “Enterprise clients are rightfully terrified of data leakage, and they will not authorise deep integration until we have strict data retention agreements and isolated secure enclaves in place.”
He also suggested that the industry will eventually transition away from massive, generic frontier models and toward smaller, highly verticalised language models that are specifically trained on advertising data and hosted within secure corporate environments.
Campbell agreed, saying that agencies must establish ironclad data boundaries when managing multiple competing clients within the same operational hub.
“We cannot allow the learnings or strategic insights from one brand to inadvertently train a shared model that a competitor might later access,” she warned.
A human and machine future
Despite challenges, it’s clear that this technology will continue to have a positive impact on the industry.
Machines won’t replace humans, instead, the participants envision a collaborative ecosystem where AI manages repetitive tasks such as execution, validation and data processing, and humans can add value by focusing on interpretation, emotional resonance and strategic growth.
Jochen Schlosser believes that this evolution requires a fundamental shift in professional training and mindset: “We are asking our teams to move from doing the work manually to designing the systems and guardrails that govern how the work is done automatically.”
For an industry defined by rapid changes in media formats, platforms and consumer behaviours, the agentic era poses its most significant and exciting transformation yet.
This is a genuine evolution that has, according to our roundtable participants, only just begun.
