AI made brand safety a production problem
AI has moved brand safety upstream. Communications teams now need to test creative, partnerships, and product surfaces for social permission before launch.
Brand safety is still managed as though the main danger begins when an ad lands beside harmful content. That model is now too narrow. AI-generated creative, technology partnerships and connected products can make the campaign itself the source of reputational risk before media placement enters the picture.
The uncomfortable shift for communications leaders is that a campaign can be technically compliant, clearly optional and well targeted, yet still feel illegitimate to the people receiving it. Audiences are judging whether a brand had the right to automate a cultural symbol, borrow a community's identity or turn a private product surface into inventory.
Brand safety now needs an upstream test of permission. Legal consent matters, but so do cultural permission, community acceptance and the expectations attached to a place or product.
Table of contents
Jump to section:
- Brand safety has moved inside the campaign
- Compliance is the floor not the permission
- The new risk map follows who feels imposed upon
- Put communications before the launch gate
- The hardest question is whether the brand earned the right
Brand safety has moved inside the campaign
Classic brand safety asks where an impression ran, what surrounded it and whether the inventory met an agreed standard. Those controls still matter. The rise of AI slop as a media-quality problem makes that much obvious. Low-quality pages can distort attention signals while placing brands beside content that weakens perceived quality.
But the same risk logic now applies to assets and experiences commissioned by the brand. A synthetic image can mishandle a national symbol. A celebrity partnership can import a contested technology into a fan community. A connected screen can carry a harmless prompt into a space where the owner never expected marketing to appear.
These are not primarily adjacency failures. The media may be owned, the placement may be intentional and the execution may pass every technical control. The unsafe element is the assumption embedded in the campaign: that production efficiency, contractual rights or access to a surface also creates social permission to use it.
That distinction matters because public trust in AI has not kept pace with adoption. A 2025 global study from the University of Melbourne and KPMG surveyed more than 48,000 people across 47 countries and found that 66% regularly use AI, while only 46% are willing to trust AI systems. Familiarity is not consent, and usage is not endorsement.
Once the brand itself creates the contested experience, a blocklist cannot protect it.
Compliance is the floor not the permission
Regulation is pushing more AI activity toward disclosure and traceability. The European Commission now states that certain AI-generated or manipulated images, audio and video must be clearly labelled under the EU AI Act's transparency framework. Its July 2026 guidance makes AI transparency an active operating requirement, not a distant policy debate.
That creates a necessary floor. It does not answer whether an audience will consider the use appropriate. A disclosure can tell people that an image was generated, but it cannot restore a misrepresented flag, supply missing cultural fluency or resolve an objection to the product being promoted.
The same gap appears in connected environments. In January 2025, the U.S. Federal Trade Commission alleged that General Motors and OnStar collected and sold precise geolocation and driving-behavior data without adequately notifying drivers or obtaining affirmative consent. The proposed order required clearer choice and consent around connected-vehicle data. The FTC action against GM and OnStar was about data practices, not advertising, but it sharpens the expectation every in-product marketing team now faces: access to a connected surface carries a heavier burden of explanation.
Consumers already connect data trust to commercial choice. Cisco's 2024 Consumer Privacy Survey found that 75% of respondents would not purchase from an organization they do not trust with their data. A prompt may not use sensitive data at all, yet it will be interpreted against the product's broader history of updates, subscriptions and data collection.
Compliance establishes what a brand may do. Reputation is decided by whether people believe the brand should have done it there.
The new risk map follows who feels imposed upon
Communications teams can map this risk by looking for the audience that bears the campaign's hidden cost. That cost may be cultural misrepresentation, pressure to defend a partnership or the loss of a previously private space.
Production risk sits with people represented by the asset. Singapore's National Day controversies showed how easily AI efficiency can become a judgment problem. One community banner was removed after residents found distorted figures and an inaccurate national flag. The useful lesson from the AI creative backlash in Singapore is not that culturally significant work must avoid AI. It is that the people closest to the symbol will see errors as evidence of how much care the brand chose to invest.
Association risk sits with communities asked to absorb a partner's controversy. When a brand or public figure promotes an AI product, objections to labor practices, environmental impact or synthetic media can attach to the partnership regardless of the creative idea. The World Federation of Advertisers found this tension inside large advertisers. In a 2025 survey of 33 respondents from 27 multinational brands, 96% cited concerns about consumer trust and acceptance around AI influencers, even while many saw potential cost and scale benefits.
Surface risk sits with owners or users who experience a new message inside a product they already paid for. BMW's optional Spider-Man infotainment experience was described by the company as themed content rather than advertising, but some owners read the unexpected prompt as commercial intrusion. The reaction to marketing inside connected cars shows why formal opt-in at the moment of interaction may arrive too late. The surprise itself can violate the expectation.
The highest-risk audience is often not the target segment. It is the stakeholder who feels that the campaign used their identity, attention or environment without first recognizing their claim over it.
Put communications before the launch gate
Most launch processes bring communications in after the concept, partner and distribution plan have hardened. At that point, the team can prepare holding lines or adjust disclosure language, but it cannot cheaply change the decision that created the risk.
An upstream brand-safety review should examine the campaign before production and media approval. The useful questions are less about predicting every possible criticism and more about exposing where the brand is treating access as permission.
- Representation: Whose culture, likeness, labor or shared symbols are being generated or remixed, and who close to that context has authority to review the meaning?
- Association: What unresolved objections follow the technology, creator or commercial partner into the campaign, and does the audience see the partnership as a values statement?
- Surface: What did the user believe this screen, device or community was for when they joined or paid for it?
- Recourse: Can people understand, decline and reverse the experience without losing access to the underlying product or relationship?
This review changes approval ownership. Legal should determine obligations. Product should verify the experience. Creative should defend the work. Communications should test the legitimacy of the premise and identify which stakeholder would be forced to carry its cost.
The point is not to hand veto power to the loudest possible critic. It is to make the commercial trade explicit while the team can still change the partner, surface or production method. Reputation work is most valuable before a holding statement becomes necessary.
The hardest question is whether the brand earned the right
AI makes more things technically possible at lower cost. Connected products create more addressable surfaces. Fandoms and creator communities offer distribution that media plans cannot manufacture on their own. Each advantage tempts marketers to treat capability as authorization.
The brands that handle this well will not be the ones that avoid contested technology or culturally ambitious work. They will be the ones that can explain why this tool belonged in this campaign, why this partner made sense for this audience and why this message deserved to appear in this place.
That explanation cannot be invented after backlash begins. It has to be visible in the choices the team made, the expertise it involved and the control it left with the audience.
Brand safety used to protect a campaign from its surroundings. Its harder job now is to stop the campaign from becoming the thing people need protection from.
