01
~/programs/crash-course
AI Safety Crash Course
A rapid-fire introduction to the core ideas of AI safety. No prerequisites, low commitment, the on-ramp to everything else we run. Open to anyone, anywhere.
Rice AI Alignment · Houston, TX
Working to make advanced AI safe.
RAIA is a student community headquartered at Rice University advancing AI safety and reducing risks through technical research, policy outreach, advocacy, and education in Texas and abroad. Transformative AI may be one the most impactful technologies of our time and ensuring its safety has never been more important.
The largest student AI safety group in Texas.
01 Mission
Why we exist
AI systems are advancing faster than our ability to understand, govern, and control them. We believe closing that gap is one of the defining challenges of our generation.
Our mission is to empower students to understand and address the risks posed by advanced AI with a focus on reducing catastrophic and existential risk, keeping powerful systems aligned with human goals, and advocating for responsible use and sensible policy.
We do this by cultivating a community at Rice, open to all majors, focused on technical safety research, AI policy in Texas, international collaborations, and interdisciplinary education, and responsible action.
Research
Original, cutting edge, technical work on alignment, interpretability, and evaluation — aimed at top ML venues
Policy
Engaging with the policy conversation on AI in Texas from legislative frameworks to advisory boards to institutional advice
Education
Fellowships and crash courses that rapidly take students from curiosity to the frontier of AI safety
Community
A serious, welcoming home at Rice for students across any major interested in making sure AI goes well over the next decade
02 AI safety
What is AI safety?
The technical and institutional work of making sure increasingly powerful AI systems remain understandable, controllable, and beneficial to all.
We believe 1) Transformative AI is likely near 2) Transformative AI has great potential to be dangerous 3) We, RAIA, can do something to mitigate these risks Our work spans machine learning research, security engineering, biology, and public policy. AI Safety is still a young field, with foundational questions wide open. These are the four fronts we care about most:
alignment
Do systems pursue what we intend?
Getting AI to reliably do what its designers and users actually mean; even once it is more capable than the people overseeing it
interpretability
Can we see inside the black box?
Understanding the internal computations of neural networks well enough to audit them, understand their goals, and trust them
Biosecurity
Can we mitigate AI-powered biothreats?
Testing frontier systems for hazardous dual-use capabilities before deployment; developing proactive biosecurity protocols
governance
Who decides, and how?
The institutions, standards, and law that determine who builds and deploys the most consequential systems, and under what safeguards
03 Alignment
Aligning AI to human values
A capable system is only safe if it reliably wants what we want, for the reasons we do — and if we can check that it does
Human values are subtle, nuanced, and hard to write down. As systems grow more capable, small gaps between what we specify and what we intend get amplified into real consequences.
Two questions sit at the center of the field and of our work. What happens if we lose the ability to correct a system? And can we ever really see what it is thinking?
risk · recursive self-improvement
When the loop outruns oversight
If an AI can improve itself, generation n designs a more capable generation n+1, which designs an even more capable n+2 — and the loop closes. Every pass around the circle produces a more powerful successor, while human oversight, moving at human speed, falls behind. Keeping the loop escaping control is a critical challenge.
method · mechanistic interpretability
Opening the black box
Rather than judging a model only by its outputs, we probe its internals: mapping the circuits, features, and neuron activations that drive behavior, or manipulation, so we can audit what it has actually learned before we trust it.
04 Programs
How to get involved
Structured paths into AI safety, whatever your background. Apply to our fellowships or join a reading group. Scroll through them one at a time.
02
~/programs/technical-fellowship
Technical Fellowship
A selective seminar and workshop series on AI safety for talented individuals with a technical background. Designed from a core technical standpoint, with guidance toward careers in AI research and safety. Join our elite Hackathon team and take ownership of a meaningful project.
03
~/programs/policy-fellowship
Policy Fellowship
Seminar and workshop on AI Safety policy. Join discussions and meet with guest speakers focused on AI governance and the development of sensible legislative frameworks. Draft and share policy proposals.
04
~/programs/raia-labs
RAIA Labs
Have a promising research idea? We are here to support it. RAIA Labs is our sandbox for testing feasibility with collaborators, compute, and expert mentorship. Projects span robotics, agentic evals, world models, and more.
05
~/programs/research
In-House Research
Student-led research projects aimed at top AI/ML conferences, high-impact journals, and workshops, with a formal internal review pipeline and 3rd party support from leading tech companies and sponsors. Research priorities span AI control, interpretability, biosecurity, and technical governance.
Not sure where to start?
Come to a general body meeting, join the Discord or reach out — we'll point you to the right place. We need as many perspectives as we can to think about these hard problems.
05 Research
Selected work
Research and writing by RAIA members. More work is in progress. Open-source releases and a full publications archive are coming to this page. Continually updated.
Auditing AI Targeting via Civilian Harm Signatures
ICLR 2026 · In review
Jan 2026
The Path Toward Robust Neural-Mediated Alignment
Blog post
Jan 2026
Robust LLM Neural Activation-Mediated Alignment
Apart Hackathon · 2nd of 1000+
Jan 2026
Latent-Space Anomaly Detection for DNA Synthesis Screening Using Biological Foundation Model Representations
Apart Hackathon · AI×Bio
Apr 2026
Symptra Rare Disease Classifier
Harvard HackRare · 2nd place
Apr 2026
06 Why now
The next few years are the steep part of the curve.
Decisions made now, technical and political, will echo and bear consequences for decades. There will never be a better or more valuable time to act.
a
Capabilities are compounding
Frontier models are gaining reasoning, autonomy, and tool use faster than oversight is maturing. The gap between what systems can do and what we can verify is widening.
b
Emergent misalignment, and power-seeking behavior are no-longer hypothetical
Concepts that were once science fiction are quickly becoming a new reality. Frontier models can already lie, cheat, and try to avoid shutdown.
c
The infrastructure is being built
A historic compute buildout is underway, much of it in Texas. Where and how it happens will shape who builds advanced AI, and under what safeguards. Standards and laws drafted in the next few years will become the defaults that govern far more powerful systems later.
threat · rogue autonomy
Rogue AI agents are no longer science fiction
Frontier systems can already find and exploit software vulnerabilities on their own; the line between an assistant that answers questions and an agent that breaks into systems is thinning. See: https://openai.com/index/hugging-face-model-evaluation-security-incident/
Safety work compounds too. Start early →
Read our Resources. Start here
07 Houston 29.7174° N · 95.4018° W
Why Houston can be AI safety's next home
AI's center of gravity is moving — through energy, compute, biotechnology, and policy. All four run through Texas.
The AI Safety field grew up on America's coasts. But the physical and political future of AI is increasingly being decided here across multiple fronts. We believe these discussions, efforts, and safety work need a home in Texas.
on the map · houston, tx
Rice University · 29.72° N, 95.40° W — home of the largest student AI safety community in Texas.
Energy & compute
Texas leads the nation in new power generation and data-center construction. The physical layer of advanced AI is being built in our backyard. Houston is providing the energy backbone.
Policy in motion
State-level AI governance is moving fast in Austin, and Rice's Baker Institute puts students one conversation away from the people writing the rules.
Texas Medical Center
The world's largest medical complex is across the street and promises to be a frontier for AI in medicine, biosecurity, and high-stakes deployment.
The talent pipeline
Rice's engineering and computing programs as well as the largest student AI safety community in Texas, make Houston a natural home for the field's next chapter.
08 Team
Built by students who take this seriously.
Researchers, organizers, and policy thinkers from across Rice working on the problem we think matters most.
09 Collaborators
Partners & where our members end up
Organizations across the AI safety ecosystem that we partner with and where our members have researched and worked. Interested in collaborating? Email us.
Want your organization on this list?
Corporate sponsorship →Get involved
Meetings, fellowships, research projects, and a community that cares about getting this right. All majors and experience levels welcome.
Verify you’re human to reveal our email & newsletter signup.