Applications open
Security is
the bottleneck.
A funded fellowship for researchers working on security for artificial superintelligence. Three or six months inside a frontier AI security lab, solving the most important problems in AI security.
ASI Preparedness
By 2027, superintelligent AI systems will be capable of autonomous cyber operations, self-directed research, and real-world action. The work to secure AI must be done now, so that the world is prepared for ASI.
ASIP funds researchers to work on that problem directly, embedded with a team doing it full time. Fellows take on a defined problem, work alongside staff researchers, and publish what can be published.
Solving the most important problems in ASI security.
Model weight security
Building defenses that let AI model weights withstand sophisticated nation-state adversaries.
Agent security
Containing and controlling autonomous agents with real-world capabilities.
Offensive cyber evals
Measuring what models can actually do against real and simulated targets.
AI cyber operations
How autonomous systems will power both defenders and attackers.
Red teaming
Adversarial work against deployed models and the systems around them.
Cyberswarms
Coordination for swarm-scale nation-state cyberdefense.
Why now
Capability is compounding faster than the defenses around it. The window to get security right closes first.
The program
Training
An intense technical training in frontier AI security.
Problem scoping
Fellows scope their problem with the team and create a plan of attack.
Research
Focused research, with speakers and visits from frontier AI lab security leaders, national security leaders, and cyber operators.
Who should apply
We care about demonstrated ability on hard technical problems, not credentials. Fellows may come to this work from academia, industry, or government.
Security engineers and red teamers
Offensive or defensive practitioners who want to point that skill at frontier AI systems.
Late-stage PhDs and postdocs
Academics in ML, systems, cryptography or security, who want to solve high-stakes problems.
National security practitioners
Operators who have worked the offensive or defensive side and know how adversaries think.
Infrastructure and platform engineers
Engineers who build and harden the frontier AI systems like GPU, HPC, and cloud systems.
Who runs it
ASIP is run by a stealth frontier AI security lab in San Francisco solving problems in AI and national security. Fellows work with technical researchers inside the lab and gain access to secured research infrastructure during the fellowship.
Individual backers — the organizations listed are not affiliated with ASIP.
Applications open
Apply to ASIP
Deadline: September 2, 2026
Expect a reply within two weeks, and a technical conversation if there is a fit.
- Duration
- 3 or 6 months
- Stipend
- $9,000 / month + compute
- Location
- Hybrid
Application received.
We read every submission. You will hear from us within two weeks.