Source-Integrated Generative-content Attribution and Labeling
Watermarking, fingerprinting, deepfake detection, and provenance benchmarks, across all carriers.
11:59 PM Anywhere on Earth
Buena Vista, FL. Day TBA
In person
Provenance has two sides. Post-hoc detection judges content after it circulates. Source-side marking embeds a recoverable signal at creation time and reads it back later. SIGNAL is about the second: marking as an engineering problem, measured honestly.
The California AI Transparency Act, operative since August 2, 2026, requires generative providers to embed a latent disclosure that is permanent or extraordinarily difficult to remove. No benchmark defines that phrase. Making it measurable is what SIGNAL is organized around.
The research is currently split into separate literatures for image, audio, video, model, and sequence marking, even though the attacks generalize across carriers. SIGNAL treats them as one applied problem and brings the communities to one room.
Image and video watermarking, in-generation marking for diffusion models, robustness to compression, cropping, regeneration, and screenshot re-capture.
Watermarking for TTS and voice cloning, survival under codecs, re-recording, and time-stretching.
Watermarking and fingerprinting neural networks, IP attribution, generator attribution from intrinsic artifacts, robustness to fine-tuning, distillation, and quantization.
Marking DNA sequences and engineered organisms, encoding under mutation, recombination, and sequencing noise.
Watermark removal, forgery, and spoofing. The arms race as a first-class research object.
Unified capacity, robustness, and imperceptibility formulations. Whether a threat model or evaluation protocol transfers across carriers.
C2PA and Content Credentials, interoperability, deployment at platform scale.
Shared test suites, attack batteries, and reporting standards for robustness claims.
The list above is a guide, not a fence. If your work touches watermarking, fingerprinting, provenance, or the benchmarks that measure them, in any carrier, it belongs here. When in doubt, submit or write to us.
All deadlines are 11:59 PM Anywhere on Earth unless stated otherwise.
Half day, in person. The agenda will be adjusted as accepted papers and speaker titles are confirmed.
Talks span the carriers the call covers, from visual media to biological substrates and the policy layer above them. The final lineup may be adjusted, and confirmed names are posted here with talk titles.
On marking DNA, and what mutation-tolerant encoding demands that media channels do not.
Biological substratesOn how disclosure requirements are written, enforced, and assessed.
PolicyWe are inviting additional speakers on audio and speech, media forensics, and language models. Names and talk titles will be posted here as they confirm.
The committee is still growing. If you would like to review, write to us.
Staff Scientist and Genome Analysis Group Lead at the DOE Joint Genome Institute. Author of genomic foundation-model work including DNABERT-S and GenomeOcean.
Senior co-organizer · BiologyWorks on artificial intelligence and image processing, watermarking, knowledge representation, and distributed computing at scale.
Organizer · Visual trackWatermarking, content authenticity, and deepfake detection, with agent-based simulation of how generative media propagates online.
Organizer · Primary contactComputational biologist building pipelines that read signal out of noisy sequencing data, where mutation corrupts the record as no media channel does.
Program committee · SequencesResearch Fellow and instructor of CS 389: Internet Censorship. Platform governance and the US regulatory landscape for synthetic media.
Program committee · PolicyHuman factors and the usability of disclosures. How a provenance signal is read, trusted, and acted on by the people it is meant to protect.
Program committee · Human factorsFor questions about submissions, the awards, or the workshop generally, write to the organizing committee.
SIGNAL is organized with support from Phi Lab Foundation, a non-profit for public-benefit AI research and open-source ecosystems, which funds the awards, speaker travel, and long-term hosting of all workshop artifacts.