Nonprofit AI safety research · Canada / Remote
Lida Safety Research is a technical AI safety lab working on white-box interpretability, AI security, and world-scale simulation for governance — and on getting those findings in front of the people who set policy.
Fig. 1 · Sky chart, RA 0h–24hThe sky turns, predictably. The futures of AI don’t — yet.
Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.
Signed by 600+ AI researchers and executives
Focus areas
What does a model do when no one is watching, and what would it take to notice in time? We attack that from the weights, from the attack surface, and from the world the systems will be deployed into.
Building model organisms of scheming and sandbagging, then reading the weights for the structure those behaviours share. If deception has a signature, we want a detector that does not depend on catching a lie in the output.
Fragmentation attacks split a prohibited task into pieces that each look harmless. We study how those attacks compose across calls and models, and what a defence that reassembles intent has to look like.
Policy proposals are argued about far more often than they are tested. We build agentic simulations of countries, labs and markets so a proposal can be run before it is written into law.
LidaSim
Wargaming
LidaSim puts agentic AI and human players into the same global-scale simulation — states, labs, supply chains and the incentives that connect them — so a scenario can be played out rather than only described.
Step 1
Set the scenario
A capability jump, a leaked model, a race between labs — the starting conditions we want to stress.
Step 2
Populate the world
AI agents play countries, companies and markets; human players take seats alongside them.
Step 3
Play it out, many times
Run the same world under different policies and starting conditions instead of arguing one story.
Step 4
Measure what a policy buys
Compare outcomes with and without the intervention to see how much risk it actually removes.
It is where our forecasting work becomes something a policymaker can sit down and lose at — and the same engine drives our Simulating AI Policies fellowship project this term.
Selected results
Fellowship
Fall 2026
Three fellowship projects are active this term, run with the SPAR Fall 2026 cohort — 24 fellows and 3 mentors across 18 hours of time zones. Each project is meant to end in something usable: a detector, a prediction, or a simulation others can run.
We create model organisms for several distinct types of scheming, then apply modern interpretability techniques to ask whether they share a common structural pattern — one signature a monitor could look for instead of a taxonomy it has to enumerate.
9 fellows
Can models collude without communicating? We test coordination across model families and fine-tunes, and use interpretability to predict, in advance, which pairs are likely to find the same focal point.
6 fellows
An agentic simulation of the world — countries, companies, and the pressures between them — used to evaluate how much a given policy actually reduces AI risk, and where it backfires.
9 fellows
Steering models away from sandbagging by grafting a reference activation into the forward pass.
Steering away from sandbagging without paired honest and malicious input–output examples.
Defences against fragmented misuse attacks, with mentored SPAR projects and hackathon placements behind them.
Working notes and longer write-ups go up on blog.lidasafety.org.
Cohort
SPAR Fall 2026
Twenty-four fellows and three mentors, split across the three projects above. Doctoral and master’s candidates and postdocs at UCLA, Northwestern, NYU, Deakin and IDEAS NCBR; graduates of MIT, UC Berkeley, Minerva and UW–Madison; and working engineers from Amex, Instacart, Netflix, Luminance, Applied Intuition and Bain — spread over eighteen hours of time zones, from the Pacific coast to Australia.
9 fellows
Lead author of “The Secret Agenda” on strategic deception in LLMs
UTC−8
AI/ML Engineer at American Express
New York
AI Engineer · evals, control, red-teaming
Italy
Software Engineer · AI control and deception
Netherlands
AI/ML Engineer · interpretability, fairness, geometry ML
Houston
Postdoctoral researcher · mech interp, AI security, red teaming
Australia
Incoming software engineer at Netflix · CS, UW–Madison
UTC−8
Postdoctoral Researcher at Deakin University
Australia
Fellow
UTC−6
6 fellows
Mathematics PhD student at Northwestern; random matrix theory at UC Berkeley
Chicago
PhD candidate at IDEAS NCBR · AI security and graph neural networks
Poland
Senior ML Engineer at Instacart · UC Berkeley School of Information
UTC−8
Astha MehtaLida team
Lead ML engineer at Bain & Company
London
Cedric LamLida team
MS student at NYU · Trust & Safety background
New York
Fellow
UTC−6
9 fellows
AI engineer at Luminance on production LLM systems and model trust · MIT
New York
Engineer at Applied Intuition
UTC−8
Independent AI researcher · evals, agentic systems, policy-impact forecasting
India
Melanie BuiLida team
AI Engineer and AI safety researcher · AI control, compute governance
Vietnam
Tra My (Chiffon) NguyenLida team
Researcher at Lida Safety · ML & Statistics, Minerva University
Vietnam
Zachary SchlosserLida team
Led the AI program for a US policy accelerator
New York
Yosra Kazemi
Independent AI researcher · causal inference, reinforcement learning
Toronto
Fellow
UTC+10
Phuc-Nguyen Nguyen
Fellow
Vietnam
Team
A small nonprofit group of researchers and engineers, deliberately spread across countries and languages — because AI safety work that only happens in English, in two cities, is not finished work.
David Williams-King
Co-founder · Executive Director
AI safety researcher and research manager focused on x-risk reduction. He works at the AI safety fellowship ERA and runs AI security events like FAST in Singapore, and enjoys mentoring people new to the field.
Linh Le
Co-founder · Research Director
Independent AI safety researcher affiliated with the Oxford AI Governance Initiative; previously Mila and the University of Technology Sydney. Dreams of running simulations of AI policies to avoid catastrophic outcomes for the world.
Researchers, engineers and fellows
Hong Kiat Tan
Researcher · SPAR mentor
Finishing a PhD in Mathematics at UCLA. Your average mech interp enjoyer — developing and applying mechanistic interpretability to safety and alignment.
Melanie Bui
Researcher
AI safety and governance researcher, focused on coordination failures and governance solutions for middle powers navigating the AI transition.
Zachary Schlosser
Researcher
Researcher, educator and convener on AI policy. Previously led the AI program for a US policy accelerator; now builds the field around formal methods for AGI uncontainability.
Cedric Lam
Researcher
AI safety researcher from NYU with a Trust & Safety background. Works on cybersecurity and AI governance for creative spaces, especially AI-generated content.
My (Chiffon) Nguyen
Researcher
Works on keeping AI safe and empowering — reducing loss of control and disempowerment, multi-agent systems, multilingual AI and AI for good.
Arthur Collé
Agent backend engineer
Distributed systems engineer. Spent the last three years building production multi-agent systems and doing autonomous agent research. Based in Washington, DC.
Astha Mehta
SPAR fellow
Lead ML engineer at Bain & Company, working on AI safety and security.
Niruthiha Selvanayagam
SPAR fellow
PhD student at École de technologie supérieure (ÉTS Montréal), working on AI security.
Get involved
One inbox, read by both directors. Tell us which of these you have in mind and we will route it.
Interpretability, AI security or simulation — if your question overlaps with ours, we would rather work on it together than in parallel.
Propose a collaboration →Donations go through our fiscal sponsor, Anti Entropy, a 501(c)(3) (EIN 88-0967420), and are tax-deductible in the United States.
Ask about donating →Fellows join each term through SPAR; experienced researchers can also mentor a project or apply to us directly.
Write to us about a cohort →Based
Canada / Remote
Fiscal sponsor
Anti Entropy, 501(c)(3)