NTT DATA exposed critical vulnerabilities in AI watermarking systems in December 2025, just as the EU AI Act's watermarking mandate entered force. The timing has accelerated research into both defensive and offensive watermarking techniques across European institutions.
"With the EU AI Act mandating watermarking, the topic has become increasingly urgent," said Shayleen Reynolds, a security researcher tracking the developments. Reynolds warned that "the watermarking vulnerability findings expose a foundational vulnerability in today's AI trust."
The EU AI Act requires watermarking on AI-generated images, audio, and video content to combat misinformation. Companies deploying general-purpose AI models in Europe must implement detectable watermarks by mid-2026 under the regulation's transparency provisions.
Security researchers are now racing to identify weaknesses before malicious actors exploit them at scale. NTT DATA's findings demonstrated that watermarks could be removed or manipulated without degrading content quality, undermining the regulation's core premise.
Patent filings for watermarking technologies have increased 40% in early 2026 compared to the same period in 2025, according to European Patent Office data. Academic publications on adversarial watermarking attacks jumped similarly, with conferences reporting double the submissions on the topic.
The pattern mirrors previous security arms races. Digital rights management systems faced similar cycles of implementation and circumvention in the 2000s. Cryptographic watermarking techniques now compete with neural network-based removal methods.
European AI companies face compliance deadlines while the underlying technology remains contested. Microsoft, Google, and Meta are deploying proprietary watermarking systems, each with different vulnerability profiles. Standardization efforts through CEN-CENELEC aim to establish common frameworks by late 2026.
The security community expects vulnerability disclosures to accelerate through 2026 as researchers probe deployed systems. Bug bounty programs specific to watermarking have launched at major AI labs, offering rewards up to €100,000 for critical findings.
Industry observers note the regulation created urgency without ensuring technical maturity. "We're mandating a technology that's still evolving," one Brussels-based policy advisor said. The result is simultaneous deployment and security research on an unprecedented scale.