Deepfake food fraud threatens global brands

Key takeaways:

• AI-generated images, fake complaint letters and manipulated food photographs are creating a new form of food fraud that existing legislation was never designed to address.
• Consumers increasingly struggle to distinguish AI-generated food content from genuine images and reviews, making fraudulent claims harder for businesses to identify and challenge.
• As AI tools become more sophisticated, food brands face growing reputational, financial and regulatory risks from digital evidence that may not be real.

A photograph showing mould on a muffin. Another appearing to reveal glass embedded in a cookie. A third depicting an undercooked pastry allegedly delivered to a customer.

In the past, such images would have been treated as compelling evidence. Today, food safety experts warn they could have been created in seconds using freely available AI tools.

UK food safety consultancy Food Alert says it’s already handling cases involving manipulated food images and AI-written complaints designed to secure refunds, compensation or regulatory intervention. While confirmed incidents remain relatively rare, experts argue the conditions for widespread abuse are rapidly falling into place as AI tools become cheaper, faster and more convincing.

The scale of AI adoption is accelerating rapidly: more than 34 million AI-generated images are now produced every day, while AI image editing has emerged as the fastest-growing software category of 2024, expanding 441% year-on-year. What once required specialist editing skills can now be achieved with a free account and a simple text prompt.

The risk extends well beyond refund fraud. As brands increase their reliance on e-commerce, quick-commerce and third-party delivery platforms, a single AI-generated image alleging contamination or poor quality can spread across social media long before a business has the opportunity to investigate the claim. In an industry built on trust, that creates a potentially costly new vulnerability.

Consumers can’t spot the difference

The industry’s ability to challenge fraudulent complaints is being undermined by the uncomfortable reality that consumers are becoming increasingly unable to distinguish AI-generated food images from genuine photographs.

Food Alert points to a 2025 study involving food photographer David Robson, who recreated dishes both through traditional photography and using OpenAI’s DALL-E image generator. When shown the images, 73% of participants failed to identify an AI-generated margherita pizza, while 69% failed to spot an AI-generated bowl of cacio e pepe and 66% couldn’t distinguish an AI-generated croissant from a real one.

Trust is becoming a broader problem. Researchers at the Yale School of Management recently found consumers struggled to distinguish between genuine and AI-generated restaurant reviews, raising questions about the reliability of the online feedback that increasingly shapes food purchasing decisions.

Food Alert isn’t alone in raising the alarm. Los Angeles-based TruthScan, a fraud-detection company specialising in digital commerce, has reported growing use of AI-generated imagery in refund disputes, where manipulated photos are allegedly being used to support claims for products that were never damaged. The firm argues that manual review systems are increasingly struggling to keep pace with AI-generated evidence.

Similar concerns have been raised by Caroline Green, a retail and supply chain lawyer at UK law firm Browne Jacobson. Speaking to The Times earlier this year, Green warned that fraudsters are becoming increasingly sophisticated through AI-powered image editing tools. The report described manipulated images showing undercooked burgers, mould, foreign objects and other apparent defects being used to pursue refunds from delivery platforms including Deliveroo, Uber Eats and Just Eat, reinforcing concerns that AI-assisted refund fraud is becoming more difficult for businesses to identify.

“A customer service team processing dozens of complaints a day has far less time to examine a single image than those test subjects did,” says Alasdair Dean, AI lead at Food Alert.

The consultancy says two distinct forms of AI-assisted food fraud are beginning to emerge. The first involves manipulated images submitted as evidence of contamination, undercooked products or foreign body incidents. The second involves AI-generated complaint correspondence designed to pressure businesses into issuing refunds.

According to Dean, AI-assisted complaint letters frequently mimic the language of legal professionals and consumer advocates, creating a level of authority that can intimidate operators unfamiliar with food legislation.

The problem is particularly acute for businesses operating through delivery aggregators. Unlike restaurants, where staff may have witnessed food preparation and service, products sold through delivery channels often lack any independent verification point between dispatch and complaint submission. When refunds are processed automatically, merchants may be forced to absorb both the refund and the original product cost while carrying the burden of investigating the claim.

A regulatory blind spot

Food fraud legislation was never designed to deal with AI-generated evidence.

“Food fraud in the UK is defined as the deliberate and intentional substitution, addition, tampering or misrepresentation of food, ingredients or packaging for financial gain, with the intention of deceiving the business,” says Annabel Kyle, technical director at Food Alert.

“Legislation is about protecting the consumer and making sure that food businesses produce and place safe food on the market, so this type of activity is not currently covered by food safety legislation.”

Kyle argues there’s little legal oversight governing the creation of AI-generated imagery. “In the UK, there is currently nothing that governs the generation of images outside the intentional generation of sexually explicit images. This also means other people might see and hear of this type of fraud and the lack of legislation around it and carry it out for themselves.”

The regulatory gap is compounded by weaknesses in the technology itself. Food Alert highlights research by arXiv, which suggests that only 38% of AI image generators currently implement adequate watermarking, while just 18% provide sufficient deepfake labelling despite requirements introduced under the EU AI Act. Even when such safeguards exist, metadata can often be removed through simple screenshots or image exports.

That creates a growing verification problem for businesses investigating complaints. A manipulated image and a genuine image may appear identical on screen, particularly when viewed through customer service portals or social media platforms.

Kyle warns that duplicitous reports can have consequences even when they’re eventually disproved. “Fraudulent complaints escalated to local authority EHO departments, whether through AI-generated images or intimidatory written correspondence, can trigger inspections. Even where the original complaint is fabricated, an inspection may uncover unrelated issues, creating real regulatory exposure. Repeated complaints on record can also, over time, affect a business’s food hygiene rating.”

Food Alert’s own research found that 84% of councils across England, Wales and Northern Ireland charge businesses for food hygiene re-ratings, with the average fee reaching £220.00.

The reputational risk is far bigger than the refund

The greatest threat may be reputational rather than operational for food brands.

An AI-generated image alleging contamination can circulate rapidly across social media, potentially reaching thousands of consumers before a company has verified whether the complaint is genuine. Food Alert notes that 67% of consumers expect brands to disclose when AI has been used to create marketing imagery, yet there’s currently no equivalent expectation requiring consumers to disclose when images themselves have been manipulated.

Unlike a refund dispute, which may affect a single transaction, an AI-generated contamination image shared online can damage trust across multiple markets. For manufacturers operating internationally, the speed at which allegations spread often far exceeds the speed at which they can be investigated.

AI-generated food content is also increasingly being used to influence how food is presented online, although not always successfully. In 2024, Instacart removed AI-generated food images following consumer backlash, while catering platform Forkable faced criticism last year after replacing restaurant photographs with AI-generated alternatives. The incidents highlighted growing unease over authenticity and transparency in digital food marketing.

Food Alert has also encountered cases involving disgruntled employees using unrelated or historical images to support malicious complaints against bakeries. As AI image generation continues to improve, distinguishing between genuine and fabricated evidence is likely to become even more difficult.

As such, it’s urging businesses to strengthen complaint management systems, improve staff training and develop structured processes for identifying suspicious patterns across complaints. Multiple claimants using near-identical language, targeting the same products or submitting structurally similar images may indicate coordinated fraud rather than isolated incidents.

It also warns the problem will intensify before effective safeguards emerge. Forecasts suggesting a 24-fold increase in AI token consumption by 2030 point to a future in which image generation becomes faster, cheaper and more sophisticated.

That trajectory suggests AI-generated food imagery will become more realistic, more accessible and more difficult to detect. For businesses still relying heavily on photographic evidence during complaint investigations, the question is no longer whether the technology will improve, but whether existing processes can keep pace.

“The businesses best positioned to navigate the next three to five years will not necessarily be those with the most sophisticated AI detection technology,” says Dean.

“Instead, they will be the ones armed with clean complaint data, well-trained staff, and robust systems designed to make coordinated fraud visible before it scales.”

Study:

Bram Rijsbosch, Gijs van Dijck, Konrad Kollnig. Missing the Mark: Adoption of Watermarking for Generative AI Systems in Practice and Implications under the new EU AI Act. arXiv:2503.18156 [cs.CY] https://doi.org/10.48550/arXiv.2503.18156