The 3rd Edition Workshop on

Synthetic & Adversarial ForEnsics
(SAFE)

Presented at AAAI 2027

February 22 or 23, 2027 · Full Day (day TBA)

at Palais des Congrès de Montréal, Montréal, Canada

Submit Paper

Call for Papers

Topics: Synthetic media forensics, adversarial machine learning, watermarking and provenance, deepfake detection, MLLM robustness, and red teaming across vision, audio, text, and multimodal content

We invite original research contributions in (but not limited to) the following areas:

  • Generative models (diffusion, GANs, VLMs, TTS/VC) for creating synthetic media
  • Detection, attribution, and robustness assessment of AI-generated content across modalities (image, video, speech, text) and generative model types
  • Multimodal large language model (MLLM) robustness to synthetic and adversarial inputs
  • Adversarial attacks across modalities, including evasion perturbations, poisoning, uncertainty attacks, and joint synthetic-adversarial techniques
  • Red teaming of generative models, including jailbreaks, prompt injection, and safety evaluations
  • Passive forensics, including manipulation-artifact detection, GAN fingerprints, and model attribution
  • Active defenses, including synthetic text detection, watermarking, fingerprinting, and provenance tracking of digital media
  • Deepfake detection, including voice-clone attribution and cross-modal consistency verification
  • Benchmarks, datasets, and evaluation metrics for detection, attribution, and robustness
  • Theoretical foundations, including information-theoretic limits of detectability and watermarking, certified guarantees, and generalization and transferability across models
  • Ethical, legal, and societal implications, including policy frameworks for responsible AI, misinformation and manipulation risks, and trust in digital media

We invite full papers (6–8 pages, excluding references) formatted with the AAAI-27 author kit. Accepted papers will be presented orally or as posters. Publication in the SAFE Workshop Proceedings on PMLR (pending confirmation) is optional: authors can choose to archive their paper there or keep it non-archival. At least one author of each accepted paper must present in person in Montréal, per AAAI policy.

All submissions undergo double-blind peer review on OpenReview, following the AAAI Code of Professional Ethics and Conduct and the AAAI Publications Policies and Guidelines. A dedicated AI-Content Integrity Area Chair screens submissions for machine-generated content; flagged papers receive additional human review rather than automatic rejection.

Important Dates

Submission Deadline
(conversion tool)
Notification of Acceptance
Camera-ready Deadline
(conversion tool)
Workshop
February 22 or 23, 2027 · Full Day

Keynote Speakers

Photo of Dr. Nouha Dziri
Dr. Nouha Dziri
Cohere Labs
Photo of Dr. Pierre Fernandez
Dr. Pierre Fernandez
Meta FAIR
Photo of Prof. David Harwath
Prof. David Harwath
The University of Texas at Austin
Photo of Prof. Xin Wang
Prof. Xin Wang
National Institute of Informatics
Photo of Dr. Elie Khoury
Dr. Elie Khoury
Pindrop
Photo of Prof. Ghada Almashaqbeh
Prof. Ghada Almashaqbeh
University of Connecticut

Organizers

Dr. Josué Martínez-Martínez
MIT Lincoln Laboratory
Dr. Pooya Khorrami
MIT Lincoln Laboratory
Ms. Nicole Meng
Tufts University
Dr. Lin Zhang
Independent Researcher
Dr. Pedro Torres-Carrasquillo
MIT Lincoln Laboratory

Area Chairs

Audio / Speech
Dr. Jonas Borgström
MIT Lincoln Laboratory
Vision
Dr. Sheila Alemany Blanco
MIT Lincoln Laboratory
Multimodal
Ms. Danielle Sullivan
MIT Lincoln Laboratory
Agentic AI
Dr. Jieren Deng
Capital One
AI-Content Integrity
Dr. Giselle Zeno
MIT Lincoln Laboratory