The dramatic growth of AI-generated content is reshaping business risk faster than most organizations can adapt. Risks associated with AI-generated content include theft of intellectual property, erosion of consumer trust, and deepfake attacks that are occurring with alarming regularity.
According to Gartner’s 2025 Cybersecurity Leader Survey of nearly 500 senior business executives, 62% of organizations have experienced at least one deepfake attack in the past 12 months, 43% reported at least one deepfake voice call incident, and 37% have experienced a deepfake on a video call.
The cost of deepfakes can be high. Global engineering firm Arup lost $25.6 million in January 2024 due to a single incident in which a finance employee participated in a video-like meeting with the CFO and other colleagues that was entirely an AI-generated deepfake.
The inconvenient truth is that AI-generated counterfeits are becoming increasingly difficult to detect.
AI watermarking is a potential solution to provide a degree of verifiable authenticity to AI-generated content. The use of watermarks is gaining momentum as new regulations such as the EU AI Act and the California AI Transparency Act include requirements for watermark use.
How AI watermarks work
Watermarks are not a new concept. It has been used on banknotes, stamps and official documents for centuries to prove authenticity and limit the risk of counterfeiting. In the digital age, this proven approach is being adapted to new media.
AI watermarking embeds a unique, recognizable signal into content during or after generation, creating a digital signature that verifies authenticity without compromising quality. This process involves two stages: encoding the watermark during model training and detecting the watermark after content generation, and requires two main approaches:
Metadatabase system. The Coalition for Content Provenance and Authenticity (C2PA) oversees open source technical standards used to verify the origins and subsequent history of media. It encrypts and embeds information about who created the content, when, where, and with which tools. This data is transferred along with the file and can be verified using free tools, making any tampering immediately apparent. The C2PA coalition currently includes Adobe, BBC, Google, Meta, Microsoft, OpenAI, Publicis Groupe, Sony, and Trupic.
pattern-based system. These watermarks modify the content in ways that are imperceptible to humans but detectable by algorithms. Google DeepMind’s SynthID subtly biases the words AI selects during text generation, creating statistical patterns that are invisible to readers but discernible through analysis. SynthID watermarks content across text, images, audio, and video.
The strongest arguments for adopting Watermark are strategic preparation, building operational experience, and governance memory.
nick caleCisco CX Engineering Principal Engineer
Business benefits of watermarks
Watermarks have a variety of potential business benefits.
Regulatory compliance and risk management
EU AI law includes mandatory AI watermarking for the first time. By August 2026, AI-generated output must be marked with machine-readable markings. The penalty for non-use of these amounts to €15 million or 3% of the global turnover of the previous financial year, whichever is higher. California’s AI Transparency Act requires AI providers with more than 1 million monthly users to implement invisible watermarks by January 2026.
According to Nik Kale, principal engineer at Cisco CX Engineering, complying with these new regulations is the strongest rationale for acting now. He said companies that implement watermarking will be better prepared operationally, even if the underlying technology remains incomplete.
“The strongest rationale for implementing watermarking is strategic preparation, building operational experience, and governance muscle memory in advance of future regulatory and policy requirements,” Kale said.
Jean-Claude Renaud, CEO of AI content detection technology provider Winston AI, had a similar view. “Watermarks make sense not as a silver bullet, but as part of a broader trust and governance stack,” he explained. “Early adoption allows businesses to prepare for regulations, partner requirements, and future provenance standards without having to panic later.”
Reliability and intellectual property protection
Watermarks provide verification infrastructure as the volume of deepfake incidents increases. Companies that seek to trace the origins of AI-generated content demonstrate technology maturity to customers, partners and regulators, Kale said. He added: “While this is no guarantee of security, it does demonstrate seriousness and preparedness when used as part of a broader governance program.”
Customer trust and transparency
Companies using AI-generated content should be concerned that a lack of transparency can erode trust. Watermarks help reduce this risk. “For customers, regulators, and business buyers, watermarks signal intent and show that they are serious about content transparency, even if the tools aren’t perfect yet,” Renaud said.
Sector-specific benefits
Early adopters who build their watermarking infrastructure ahead of regulatory deadlines will gain compliance benefits and brand trust status.
Watermarks are becoming even more important with the use of AI in education and healthcare. Tiffany Masson, founder of AI consultancy Falcovia, said universities could use this to alleviate concerns about students no longer creating original content for courses. He said there will continue to be pressure to demonstrate that systems are in place with relevant policies and procedures. In healthcare, transparency is important for healthcare providers who use AI-generated recommendations to ensure ethical healthcare for patients, she noted.
Differentiation from competitors
Early adopters who build their watermarking infrastructure ahead of regulatory deadlines will gain compliance benefits and brand trust status. This approach demonstrates proactive governance to customers and business partners.
Challenges and limitations of AI watermarking
Although watermarks have important benefits, current technology faces limitations that businesses must understand.
reliability issues
Our track record shows that reliability issues persist. OpenAI launched an AI text detector for ChatGPT in January 2023, but shut it down six months later due to poor accuracy. This failure highlights a fundamental challenge. Watermarks are often easily removed or degraded, especially through everyday content workflows.
“Most watermarking systems tolerate light compression and simple re-encoding well,” says Renaud. “When you introduce cropping, resizing, screenshots, format hopping, or copy-and-paste workflows, reliability deteriorates rapidly.”
An even bigger problem, Renaud added, is that watermarking only works if the entire pipeline is working together. The watermark disappears when the metadata is removed and the content is flattened or re-rendered in a single step.
Companies need to set realistic expectations, Kale said. He explained, “From an enterprise risk management perspective, organizations need to consider watermarks as deterrents and signals of intent, rather than reliable methods of acting as tamper-proof or forensic evidence.”
Risk of false positives
Malicious attackers can add watermarks to authentic human-authored content, casting doubt on its legitimacy. A pattern that mimics a watermark can be created by chance, potentially leading to false accusations. These risks complicate business decisions regarding content verification and dispute resolution.
Barriers to adoption
Technical vulnerabilities have limited widespread adoption of watermarking technology. Beyond that, Renaud says the following key obstacles are preventing widespread adoption:
Fragmentation. There is no universal standard that works across models, platforms, and downstream tools. A watermark applied on one system may be unreadable on another system.
false confidence. Some companies consider watermarks to be equivalent to protection and traceability, but in reality, watermarks can be easily removed, intentionally or unintentionally. This way of thinking creates a dangerous gap between perception and reality.
Lack of immediate ROI. Watermarks are primarily defensive. By itself, it does not increase revenue, improve performance, or improve the user experience.
The future of AI watermarking
Watermarks are rapidly evolving. Breakthrough technologies like watermark ensembles allow multiple watermarks to coexist without overwriting each other, creating stronger provenance chains. Zero-knowledge proof systems allow verification of detection algorithms without exposing them to the public.
But the arms race is underway. Researchers have discovered that watermarks can be removed through sophisticated attacks. Commercial bypass services advertise high success rates. A realistic goal, experts say, is not to achieve perfect detection, but to raise barriers.
Alongside watermarking, alternative approaches are gaining attention. Post-mortem detection tools analyze statistical patterns in your content, with varying degrees of accuracy. The consensus view favors a layered defense that combines watermarking with metadata standards, detection tools, and organizational protocols.
IT leaders should consider the following steps to leverage the potential of AI watermarking.
Audit all AI systems and outputs and categorize them by regulatory risk category.
Adopt C2PA content credentials for published media.
Establish internal policies that require disclosure of AI-generated content.
Join industry standards bodies to influence evolving requirements.
With further regulations on the horizon, including watermarking requirements, time is of the essence for businesses to be ready to comply.
“Companies that move now will learn, be operationally ready and gain credibility,” Renaud said. “In fact, it’s often more valuable than waiting for a technically perfect system that may never be fully completed.”
Sean Michael Kerner is an IT consultant, technology enthusiast, and tinkerer. He is known for pulling Token Ring, configuring NetWare, and compiling his own Linux kernel. He consults with industry and media organizations on technology issues.
