UltraGuard watermarks face videos in real time using ultrasound biometrics
Smartphone ultrasound and video motion are fused into watermarks meant to survive forgery and editing.
Stories per day · 30 days
5 RSS
Relatedattribution3audio2copyright2evasion2llm-agents2provenance2agent-harness1agentmark1agent-security1auditing1backward-compatibility1batch-normalization1biometrics1black-box1black-box-detection1claude-code1code-generation1codellama1
54 stories, newest first · one card per story
Stories tagged #watermarking
SkillCODER fingerprints LLM agent skills for copyright and buyer attribution
Across three domains, six models and three frameworks, owner verification reached 99.2% and buyer Top-1 attribution 99.3%.
Logit Laundering evades data-use auditing in large language models
Logit Laundering manipulates model outputs at inference time using logit reshaping to suppress watermark signals while preserving model utility on downstream tasks.
NotaryMark multi-bit watermarking for T2I dataset auditing
NotaryMark aligns watermarks with diffusion latent denoising for robust black-box auditing across datasets and models.
Dynamic watermarking detects replay attacks in vision controllers
RL policy balances watermark energy against control distortion on UR5 robot.
Averaging two audio copies breaks traceability in personalized watermarking systems
Averaging only two personalized audio copies produces tracing-failure rates of 87.7-99.7% while largely preserving audio quality.
ScrambleMark evades speech watermarking via harmonic scrambling
ScrambleMark achieves 97% evasion on commercial platforms while preserving speech naturalness at 4.2/5.
SHield watermarks 4D Gaussian Splatting files
SHield embeds payloads in 4DGS via SH coefficient perturbation for decoder-free ownership verification.
PP-Mark provides provable watermarking for generative AI models
PP-Mark anchors statistical detection to zero-knowledge proofs to ensure unforgeability against forgeries.
Training-free multi-bit image watermark resists removal and forgery
Authors report a scalable 32–128-bit payload, a 94.6% decode rate against CtrlRegen+, and 98.8% forgery detection.
AgentAnchor watermarks LLM agent logs against cross-session forgery
A trusted verifier checks a disputed trajectory without rerunning the agent, using wrong-key calibration and abstention.
Watermark Basin Alignment detects distilled LLMs without model weights
At a 0.1% false-positive rate it detected 79% of distilled models at n-gram=1, versus 29% for a token baseline.
Homoglyph watermarks still favor scripts used by higher-resource languages
After copy-paste strips invisible characters, detection needs about 2 English or Persian code points but 93 Lao.
SynMark watermarks LLM code via semantic coordinates, not token prefixes
After full identifier renaming, BitAcc stays at 82.27% on CodeLlama and 83.77% on DeepSeek-Coder.
HiMark adds controllable multi-bit watermarks to LLM text
With 32 bits in 300 tokens, authors report 90.01% extraction accuracy after paraphrasing while preserving text quality.
Data-free attack removes batch-norm watermarks from federated models
Tests on two federated-learning schemes strip ownership marks without training data and with little accuracy loss.
SUMMARK checks hosted LLM identity from reasoning summaries
A trusted auditor enrolls identities in advance; audit time uses only the claimed identity and the released summary, not provider logs.
BLADE improves multi-mode attribution of LLM watermarks
On OPT-1.3B and other open-weight models, BLADE beat compared detectors across modes and substitutions.
Backward Compatibility in Zero-Bit Neural Audio Watermarking
Legacy-chain training plus angular diversity loss makes neural audio watermarks backward compatible.
What Survives Self-Evolution: Provenance Marks in Copied Agent Harnesses
Provenance marks in leaked agent harnesses mostly fail when a self-evolving LLM rewrites the marked code.
SkillMarker: A KGW-Based Watermarking Method for Agent Skill Ownership Verification
SkillMarker watermarks agent skills with KGW so owners can verify copies without breaking procedural accuracy.
ECGWmark: Deep Watermarking for Electrocardiogram Signals
ECGWmark watermarks ECG signals without quantization and keeps over 90% bit accuracy under circular shifts.
WaveMark: Robust Wavelet-Domain Image Watermarking for Diffusion Models
WaveMark watermarks diffusion noise in the wavelet domain and keeps over 88% accuracy under removal attacks.
SemField: A Simple, Linear, Continuous, yet Robust Semantic Watermark
SemField embeds a training-free linear semantic watermark in sentence embeddings and beats 12 baselines under paraphrase attacks.