Linguistic security
How language itself becomes an attack surface: polyglot prompt injection, code-switching (Spanglish, Arabizi) and dialect gaps in AI guardrails.
Text-only mode is on: images and decoration are hidden. Turn it off from the Access menu.
Hello, I'm
Sahara Al-Madi
Computational linguist · AI security and governance · Founder, Linguistic Security Institute
"Language is inheritance for communities, infrastructure for policy, and evidence for science. Systems should serve all three."
A linguist building systems that are secure, accountable, and educational for the communities they serve. Founder of the Linguistic Security Institute — an open-source research initiative built on one principle: AI systems must be accountable to the communities whose languages they process.
Published at ACL 20267 talks & workshops in 2026Open-source tools & live demos
Areas of work
A linguist building systems that are secure, accountable, and educational for the communities they serve.
My work asks not only whether a system works, but for whom it works, whom it fails, and who gets to decide — across computational linguistics, cybersecurity, AI governance, conservation and space policy.
AI governance
Risk assessment frameworks and accountability protocols for language integrity across the AI lifecycle.
Indigenous data sovereignty
Community consent and control over data about their languages, lands and knowledge — from archives to Earth observation.
Multilingual NLP & evaluation
LLM evaluation, error analysis and golden-label quality for Arabic, Spanish and dialectal variation.
Language revitalization
Tools for low-resource and endangered languages, like Mitote, a Nahuatl-aware text-to-speech prototype.
Earth observation & space
Data ethics for geoscience and conservation, and long-horizon thinking for responsible space-debris prioritization.
Collaborations
Talks, research and community
Keynotes · workshops
Speaking & workshops
Conference talks and hands-on red-teaming workshops on linguistic attacks in AI, data sovereignty and language technology.
Invite me to speak →Shared tasks · papers
Research collaboration
Arabic NLP, multilingual benchmarking, bias auditing and error analysis across large language models.
See research →Data · QA · evaluation
Multilingual AI evaluation
Golden-label sets, QA frameworks and annotation guidelines for multilingual LLMs and voice AI.
See experience →Open source
The LSI community
LSI builds the frameworks, tools and conversations that hold AI accountable to the communities whose languages it processes.
Join LSI →Tell me about your project: Start an inquiry → or email us
LinkedIn · GitHub
On the horizon
Upcoming engagements
24Jan 2026
Tech Intersections · Northeastern University, Oakland
"Bridging the Cybersecurity Gap for Multilingual and Marginalized Communities"
On the intersection of language, security and equity.
16Mar 2026
GEO Indigenous Summit · Group on Earth Observations
"Who Speaks for the Data? A Linguist’s Lens on Listening to Land and Language"
Parallels between human-language AI and non-human data ethics for space and conservation communities.
4Apr 2026
BSides San Diego · San Diego State University
"Polyglot Poisoning: Bypassing AI Guardrails via the Linguistic Supply Chain"
Featured the Linguistic Firewall demo, examining vulnerabilities and defensive frameworks.
9Oct 2026
Yohaku Space Global Showcase · Space Debris Hackathon
"Seven Generations in Orbit: A Framework for Responsible Debris Prioritization"
Hackathon contribution solving Challenge 2, presented via public showcase video.
11Oct 2026
Geological Society of America — GSA Connects 2026 · Colorado Convention Center, Denver
"Indigenous Data Sovereignty in AI-Driven Earth Observation: A Framework for Equitable Geoscience and Education"
A proposed framework for Indigenous data sovereignty in geoscience and education.
16Oct 2026
Technology for Second Language Learning Conference · Iowa State University
"To Listen is to Inherit: From Grandmothers to Generative Models"
How human traditions of listening and carrying memory can inform what we build.
14Nov 2026
DEATHCon / BSides San Diego · University of San Diego
"Polyglot Poisoning: Hands-On Hunting for Linguistic Attacks in AI"
Pre-recorded virtual workshop plus a Saturday in-person, hands-on session.
Demos
Featured demos
Working prototypes from the Linguistic Security Institute, open on GitHub.
AI security · BSides San Diego 2026
Linguistic Firewall: Polyglot Poisoning
A prototype that probes LLM guardrails for "polyglot poisoning": using dialects, scripts and code-switching to slip past safety filters tuned mostly for English.
- Web interface and API on Python 3.9+
- Runs against local Ollama-compatible models
- Basis for the DEATHCon 2026 hands-on workshop
Responsible disclosure: full attack prompts are shared only with verified researchers working on defenses.
YOHAKU Space Debris Hackathon · Challenge 2
Seven Generations in Orbit
A decision tool for choosing which space-debris objects to take responsibility for first, built on seven questions grounded in the Lakota principles shared through the challenge.
- Six real debris objects, each with its web of relationships
- Compares today’s ranking with one seven generations ahead
- Refuses to rank until standing to decide is established
Community & media
Where the work shows up
LSI publishes across three channels — video, writing, and open models — each aimed at making linguistic security research accessible and actionable for the communities it serves.
YouTube · video series
The Cyber Lingo
Interviews, live workshops, red-team demos and visual explainers — making AI security and linguistic rights legible for everyone, not just practitioners.
Watch on YouTube →Writing · Substack · resources
Field notes & interviews
Written interviews, research articles and community resources on linguistic security, AI governance and language rights — published here and on Substack.
Read the blog →Hugging Face · open models & datasets
Open models & datasets
LSI's open-source models, datasets and evaluation benchmarks for multilingual NLP — Arabic, Spanish and low-resource language research, free to use and build on.
Explore on Hugging Face →More
- Languages
- English (native) · Spanish (native) · Arabic (reading, writing, linguistic analysis) · Italian (intermediate)
- linguisticsecurityinstitute@gmail.com
- linkedin.com/in/madisahara
- GitHub
- github.com/bintdamiana
- Hugging Face
- huggingface.co/madisahara
- YouTube
- The Cyber Lingo
- LSI
- Linguistic Security Institute
- CV
- Download my CV (PDF)
Writing
Notes from the field
I write about linguistic security, AI governance, and what technology owes the voices it learns from. Subscribe for occasional research updates and invitations to dialogue.
Read the blog →