Loading story…
Reverse-engineering by developer Xusheng Li found that Microsoft Paint and Photos embed a server-provisioned invisible watermark in AI-generated images. Even when Paint uses a local Stable Diffusion model on an AI PC, it sends the user's prompt and selected style to Microsoft for moderation; the service returns a modified prompt, generation ID, and a watermark identifier.
The blind watermark embeds a 16-byte GUID-derived value into image pixels and is separate from the visible, user-toggleable Copilot watermark. Microsoft also attaches C2PA Content Credentials, creating both pixel-level and metadata-based provenance records. Paint treats successful invisible-watermark insertion as mandatory and fails image generation if it cannot add one, while Photos logs the failure but still delivers the generated image.
Track how attackers are adapting to this technology.
2 events from the most recent confirmed update back to the earliest known activity.
Article 50 transparency requirements of the EU AI Act took effect, a provision the reporting suggests may relate to watermarking of AI-generated content.
Reverse-engineering researcher Xusheng Li reported that Microsoft Paint and Photos embed a server-provisioned GUID-derived invisible watermark in AI-generated images, alongside C2PA Content Credentials. The findings showed Paint fails image generation if blind-watermark insertion fails, while Photos logs the failure but still returns the image.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
Follow how adversaries are adapting to this technology, and where it touches your stack today.
3 references tracked. Mallory keeps watching after this page renders.
cyberveille.ch
Open sourcetomshardware.com
Open sourcesecurityonline.info
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
Major frontier AI providers including **Anthropic, Google, OpenAI, and Meta** have begun deploying watermarking and provenance controls on synthetic content to comply with **EU AI Act Article 50**, which requires general-purpose and generative AI systems to mark AI-generated outputs in a machine-detectable form. Anthropic said it is watermarking Claude text using a version of **Google DeepMind's SynthID**, a technique that subtly adjusts token-selection probabilities so longer passages can later be identified as AI-generated; Google has already used SynthID in Gemini outputs, while other vendors are also adopting metadata-based provenance approaches such as **C2PA**. The rollout has exposed immediate technical and operational limits. Open-source tools have appeared that can strip metadata or reduce watermark detectability, and statistical text watermarking can degrade after paraphrasing, translation, or in short outputs. The reporting also highlights false-positive risks in low-entropy material such as code and configuration files, along with a broader compliance gap between tightly managed hosted AI services and self-hosted open-weight models, where downstream operators can change decoding behavior and bypass watermarking controls.
Aug 18, 2026
The rapid advancement of generative AI technologies has introduced significant challenges in verifying the authenticity and origin of digital images and videos. Researchers at Queen’s University in Canada have investigated watermarking as a method to tag AI-generated images, enabling the verification of their origin and integrity. Watermarking systems function as comprehensive security processes, involving embedding coded messages within images either during or after their creation. These embedded signals can later be extracted and matched to cryptographic keys, confirming the provenance of the image. Some watermarking techniques are integrated directly into the generative model, increasing their resilience to common image manipulations such as compression or cropping. The effectiveness of watermarking relies on the invisibility of the mark to viewers, its durability through standard image handling, and its resistance to duplication by unauthorized parties. Early watermarking methods used signal-processing techniques, but the complexity of modern generative models has led to the adoption of deep learning-based approaches, such as encoder–decoder networks, which can automatically hide and retrieve marks. The emergence of advanced AI video generation models, such as OpenAI’s Sora2, has further complicated the landscape. Sora2, accessible through platforms like Lovart, allows users to generate photorealistic video sequences from text prompts, democratizing access to powerful content creation tools. While this technology empowers creators, it also provides malicious actors with sophisticated means to produce deceptive visual content. The widespread availability of Sora2 raises urgent concerns about the reliability of visual evidence and the potential for misuse in fraud, misinformation, and other security threats. Cybersecurity professionals are now tasked with developing and implementing robust verification frameworks to maintain digital integrity in an environment where AI-generated content is increasingly indistinguishable from reality. The need for new digital verification standards is underscored by the growing difficulty in distinguishing authentic media from AI-generated fabrications. Both watermarking and other verification technologies are being evaluated and improved to address these evolving threats. The convergence of advanced generative AI and the challenges of content verification highlights a critical area of focus for the security community. As generative models continue to evolve, so too must the methods for ensuring the authenticity and accountability of digital visual content. Ongoing research and collaboration between academia, industry, and security professionals are essential to develop effective solutions that can keep pace with the rapid innovation in AI-generated media.
Jun 29, 2026
Google DeepMind introduced **SynthID Bio**, a provenance framework that embeds detectable watermarks in AI-generated biological designs. The system includes sequence-level watermarking for protein and genomic designs, using secret-key-derived scoring during amino-acid selection, and a structure-level method that embeds a zero-bit signal in predicted protein atomic coordinates. DeepMind integrated the approach with the Evo 2 genomic model to watermark a designed bacteriophage genome; early bacterial-culture tests indicated that the watermarked phages remained functional. In published evaluations, watermarked protein binders targeting VEGF-A, PD-L1, and the SARS-CoV-2 receptor-binding domain were computationally screened and tested through in-vitro expression and binding assays. The technology is intended to help DNA-synthesis providers distinguish safeguarded AI-model outputs from unfamiliar sequences and to label AI-generated entries in databases including the Protein Data Bank, UniProt, and GenBank. Researchers acknowledged that deliberate resequencing or other tampering could remove some watermark types, particularly sequence marks, and released methods, code, data, and model weights for further study.
Oct 1, 2026