Non-archival papers
Up to 8 pages (excluding references) in the NeurIPS workshop template, submitted via OpenReview.
NLP/CORE Unranked
FLMSec 2026
Foundations of Language Model Security: Theory, Practice, and Fundamental Limits
FLMSec 2026 is a NeurIPS workshop focused on advancing the foundational research of language model security, aiming to shift from reactive attack-defense cycles to principled, secure-by-design systems. It explores formal frameworks, provable security, evaluation methodologies, and fundamental trade-offs between security and utility in LLMs.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Research areas in scope
01
Formal frameworks and definitions for LLM security
Formal frameworks and definitions for LLM securitySecure-by-design LLM system architecturesProvable security results and impossibility results for LLM-based systemsDesign of reproducible, generalizable evaluation methodologiesThe security–utility trade-off in current and future defensesModel- and system-level attacks and defenses (prompt injection, data poisoning, jailbreaks) situated within a principled security frameworkMulti-agent security and worst-case security guaranteesMemorization and privacy in language modelsCompositional security: do component-level guarantees hold when models or agents are composed into larger systems?
Policies worth checking twice
- Submissions must be non-archival and not previously published.
- Authors with a personal conflict of interest with any organizer are not permitted to submit.
- Reviewing is double-blind: all submissions must be anonymized, including code, data, and supplementary materials.
- External links (e.g., to code or data) must not reveal author identities.
- Acknowledgments must be removed at submission time.
- Self-citations must be anonymized (e.g., 'In the previous work of Jha et al. [1]...').
- Submissions must follow NeurIPS conflict-of-interest policies and submission guidelines.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026