We are excited to announce a new workshop in the Security and Privacy
Workshops series, collocated with IEEE S&P in May 2024 in San
Francisco, CA, on the topic of security for GenAI systems and
applications.

Call for Papers
Security Architectures for GenAI Systems (SAGAI) 2024
May 23, 2024, collocated with IEEE S&P
https://sites.google.com/view/sagai2024/home

Overview
Generative AI (GenAI) is quickly advancing and fast becoming a widely
deployed technology. GenAI-based systems rely on machine-learning (ML)
models trained on large amounts of data using deep-learning
techniques. As the power and flexibility of the models advance, the
architectural complexity of GenAI-based systems is advancing
too. Current architectures may combine multiple models, using
sequences of model queries to complete a task, with external (non-ML)
components leveraged to enhance the model’s operation via database
queries or API calls. These architectures may be vulnerable to a
variety of attacks that use adversarial inputs to create malicious
outputs.

This workshop invites new contributions to the broader understanding
of security for GenAI systems and applications. Contributions may
address security threats and defenses for individual models, or for
systems and architectures that may employ one or more generative ML
models as subcomponents. The workshop welcomes discussion of new GenAI
security concerns, as well as new approaches to architecting
GenAI-based systems for safety, security, and privacy.

Topics of Interest
SAGAI welcomes contributions on all aspects of safety, security, and
privacy of GenAI-based systems, including text, image, audio, video,
code, and other modalities. Topics of interest include, but are not
limited to:

Mechanisms for Safety, Security, and Privacy of GenAI

    Input sanitization, normalization, and deobfuscation
    Protections against prompt injection
    Output validation and sanitization
    Secure and private tool use
    Secure and private retrieval-augmented generation
    Secure and private multi-model/multi-agent systems
    Mechanisms for whitebox vs blackbox models
    Security and performance of on-device model training and inference
    Reliable watermarking techniques
    Attacks against GenAI safety, security, and privacy mechanisms

Security Architectures for GenAI

    In-model vs. out-of-model security approaches
    Secure sequential and parallel composition of GenAI-based systems
    Layered security for multi-agent GenAI-based systems
    Composition of provenance mechanisms (system and GenAI)
    Composition of security and privacy mechanisms (system and GenAI)
    Security uses of watermarked GenAI outputs
    Model explainability for security and privacy

Out of Scope
Because there are many other conferences and workshops on this topic,
we consider techniques for pre-training or fine-tuning the model(s)
used by a GenAI-based system, or to curate the data used in such
training or tuning, to be out of scope for the workshop. This includes
training techniques to achieve model alignment and techniques to
prevent data poisoning. However, submissions that consider alignment,
robustness, new forms of attack, and novel defenses of system
architectures that combine individual models with other components are
welcome.



We accept full-length papers of up to 10 pages, plus additional
references. To be considered, papers must be received by the
submission deadline (see Important Dates).

Paper Format
Papers must be formatted for US letter (not A4) size paper. The text
must be formatted in a two-column layout, with columns no more than
9.5 in. tall and 3.5 in. wide. The text must be in Times font,
10-point or larger, with 11-point or larger line spacing. Authors are
strongly recommended to use the latest IEEE "compsoc" conference
proceedings templates.

    LaTeX submissions must use IEEEtran.cls version 1.8b with options
    "conference,compsoc":
    \documentclass[conference,compsoc]{IEEEtran}

    Overleaf submissions must use the IEEE Demo Template for Computer
Society Co nferences.

    For other page-layout software, please follow IEEE Computer
Society's author guidelines.

Failure to adhere to the page limit and formatting requirements are
grounds for rejection without review. Submissions must be in English
and properly anonymized.

IEEE S&P’s criteria for anonymous submissions, conflicts of interest,
ethical considerations, and competing interests (all available at
https://sp2024.ieee-security.org/cfpapers.html) apply.

Presentation Form
All accepted submissions will be presented at the workshop. All papers
will be included in the IEEE workshop proceedings. One author of each
accepted paper is required to attend the workshop and present the
paper for it to be included in the proceedings.

Important Dates
Paper submissions due: February 5, 2024 AoE
Acceptance notice to authors: February 19, 2024 AoE
Publication-ready papers due: March 5, 2024 AoE
Workshop: May 23, 2024