Google’s Gemini AI Sparks Controversy by Removing Watermarks from Stock Images

Google’s ambitious AI project, Gemini 2.0, is back in the spotlight—but not for the reasons the tech giant likely hoped. A recent TechCrunch report reveals that Gemini’s experimental image generation capabilities are reportedly being used to remove watermarks from copyrighted images, including those from premium sources like Getty Images. The development has set off alarm bells across social media and among copyright advocates, sparking a new debate about AI ethics, copyright infringement, and the boundaries of responsible innovation.

As Google expands access to its Gemini 2.0 Flash model, its image editing features are drawing scrutiny for their ability to bypass digital watermark protections. While the company labels the feature “experimental” and warns it is “not for production use,” users across platforms like Reddit and X (formerly Twitter) have demonstrated that the model can reconstruct imagery beneath visible watermarks. In some cases, the model successfully reconstructs even high-quality stock photos in near-original condition.

Gemini 2.0’s Watermark Removal Ability Raises Legal and Ethical Concerns

Reports suggest that Gemini’s Flash model can remove or reduce watermark visibility, particularly on images with light or poorly embedded watermarks. However, the model struggles with semi-transparent or deeply integrated watermarks, which remain partially visible or distort the result. Despite these technical limitations, the implications are serious: AI models capable of altering copyrighted content risk violating intellectual property laws and harming content creators.

By contrast, competitors like OpenAI’s GPT-4o and Anthropic’s Claude 3.7 Sonnet enforce strong ethical boundaries. Both explicitly refuse to assist in removing watermarks or modifying copyrighted visuals, citing ethical and legal concerns. Claude even labels such activities as “unethical and potentially illegal.”

Legal Framework: Copyright Law is Clear

Under U.S. copyright law, the removal of watermarks without the permission of the copyright holder generally constitutes infringement. These marks serve not only as ownership identifiers but also as tamper-evident safeguards. The unauthorized editing of such elements, even by experimental software, can expose both developers and end users to liability, particularly if the edited content is redistributed or monetized.

Google’s Push Toward Physical AI

Amid this controversy, Google CEO Sundar Pichai continues to highlight Gemini’s broader applications. In a recent post on Twitter, he emphasized Gemini’s role in advancing robotics. According to Pichai, the new Gemini 2.0 models excel at tasks like generalization and embodied reasoning—skills essential for adaptive robots. These capabilities allow robots to interpret and respond to the world in real time using Gemini’s multimodal AI understanding.

Still, ethical questions loom large. Can a tool be trusted to drive the future of robotics and real-world applications if it also allows the alteration of copyrighted digital content? This dichotomy lies at the heart of the Gemini debate.

What Undercode Say:

The recent revelations surrounding Gemini 2.0 reflect a broader issue plaguing the AI industry: the race to push technological boundaries often outpaces ethical considerations and regulatory frameworks. While it’s understandable that developers want to showcase what’s possible, removing watermarks treads into dangerous territory—both legally and ethically.

Google’s decision to label Gemini 2.0 Flash as “experimental” doesn’t absolve responsibility. Once a model is accessible via developer tools, it’s essentially live in the hands of those who might exploit it. Whether accidental or intentional, any misuse stemming from watermark removal can damage reputations, reduce trust in AI companies, and endanger digital copyright norms.

Furthermore, this isn’t an isolated concern. Over the past year, we’ve seen growing tension between AI capabilities and content ownership, especially in creative industries. Artists, photographers, and designers already face existential threats from AI-generated replicas. Now, the possibility of models being used to scrub ownership markers compounds the issue.

From a technical standpoint, Gemini’s ability to reconstruct images beneath watermarks suggests an impressive depth of generative understanding. However, such functionality should be strictly sandboxed or restricted by design—not left to user discretion. Companies like OpenAI and Anthropic are setting ethical guardrails for a reason, and Google would do well to follow suit.

There’s also a risk to Google’s broader ambitions. As Gemini extends its reach into robotics and real-world applications, the need for public trust becomes paramount. If a model is known for editing out watermarks, will users trust it to navigate ethical boundaries in physical environments, like autonomous systems or surveillance tools?

The burden is now on Google to clarify usage limits, reinforce ethical guidelines, and demonstrate commitment to responsible innovation. This isn’t just about one feature—it’s about the message being sent to developers, users, and the global tech community. In the age of AI ubiquity, every model is more than code. It’s a statement of intent.

Fact Checker Results:

Claim: Gemini can remove watermarks from copyrighted images.

✅ Verified via user reports and TechCrunch source.

Claim: Other models like GPT-4o and Claude refuse such requests.

✅ Confirmed with public documentation and model behavior.

Claim: Watermark removal violates U.S. copyright law.

✅ Accurate under DMCA and copyright statute 17 U.S.C. § 1202.

Prediction:

If Google fails to address the watermark issue head-on, regulatory scrutiny is likely to intensify. Governments may introduce stricter AI usage rules, especially around digital copyright and content authenticity. Competing AI firms will capitalize on this moment to emphasize ethical design, potentially gaining market trust. Google’s roadmap for Gemini must now include transparency tools, in-model restrictions, and policy disclosures to remain competitive—and responsible—in an increasingly scrutinized AI landscape.