Bridge, Anchor, and Incubation

We want to bring the world’s best into AI safety. The Bridge track is an on-ramp for people bringing deep and hard-won expertise to bear on helping the future go well. If you have talent and experience that could be a key unlock for important agendas in AI safety, but face hurdles to effecting the career transition and lining up the right landing spot, we want to provide the time, resources, network, and guidance needed to facilitate your impact.

We take an expansive view of who could be a fit. Bridge track is for exceptional talent with the potential to contribute something important that are blocked by something we are well-positioned to help you overcome. We expect that common hurdles include 1) time and opportunities to build your context, strategy, and theory of change, 2) the need to build a legible record of work in AI safety, and 3) access to a high density of thinking partners and networking opportunities (such as those offered by our locations in Berkeley, DC, and London). Strong applicants will be able to articulate why MATS residency is the best path given their situation (compared to e.g. our fellowship or a career transition grant), can demonstrate expertise in an area plausibly valuable for reducing catastrophic risks of AI, and are driven by the mission.

We are open to strong applicants from any field! We are excited to hear from professors and postdocs committed to reorienting their focus, people with deep expertise in policy and governance and diplomacy, top talent in domains of immediate need (e.g. cybersecurity and cyberforensics, biosciences, epidemiology, tort and liability, safety-critical systems experts, regulatory affairs and compliance regimes, I/O psychology and labor economics, philosophy, compute governance, statistical forensics and anomaly detection etc.), and AI capabilities researchers switching into safety. We’re also happy to receive applications from existing talent in AI safety who want to substantively shift their focus.

Established academics may use a supported sabbatical, fellowship, or postdoctoral placement of up to two years to redirect their research toward AI safety, field-building, or founding a new organization. While normally taken as full-time leave, part-time arrangements may be considered on a case-by-case basis.

• Duration · 6 to 24 months
• Review Cycle · 12 weeks (first block may be 6 to 8 weeks for timing-sensitive cases)
• Key feature · Placement Desk (career development, opportunity tracking, and referral letters)
• For newcomers · Safety Orientation Week and Translation Sprint in Block 1

We think AI Safety needs more institutions that try to maximize both researcher independence and institutional resourcing and capacity. The Anchor track is intended to give people the time and opportunity to pursue long-horizon work wherever it leads, with a high concentration of thinking partners and the resources to go big. We want to empower promising research programs with support from dedicated professional compute and ops teams to help them reach their potential. We want people who would thrive in a research-institute-like context, especially those who can’t find an obvious home for their research program.

Strong applicants will be high-context AI Safety researchers excited to think expansively and ambitiously about what the future needs to go well. We are looking for outstanding examples of your ability to approach hard problems and evidence of ability to steer a research program, though this can include writing and resources outside of academic-style papers. We are especially excited by efforts to directly tackle key problems in alignment, but hope to see diverse and creative approaches to mitigating catastrophic risks of AI.

Residents will begin their residency by developing their research plan and sharpening their thinking around the topic area, with assistance from the residency team and outside advisors. Then, we encourage wide latitude to drive at the cruxes. We will help scale resourcing for promising threads and encourage travel to connect with other key experts and strategic thinkers. We are excited for anchor fellows to speak and collaborate widely, including travel to visit centers of learning in the field. We expect residents to be proactive in getting their ideas out into the community and expect the work to generally facilitate understanding and preparedness/capacity in public ways, though we don’t want some arbitrary cadence of writing to impede focus.

‍

• Duration · 1 to 2 years
• Review Cycle · 12 weeks
• Key feature · Independence Pathway Review at 12 months
• From Block 2 · Peer accountability check-ins every two weeks

We aren't convinced that all of the things we need for the future to go well have already been discovered. Incubation track is for laying the foundations of new paradigms in AI safety. Research agendas like developmental interpretability, gradient routing, and sparse autoencoders for AI interpretability were incubated at MATS, and Apollo Research, Timaeus (now part of Resolution), Simplex (now part of Astera) and many other research organizations were co-founded by MATS fellows. We want to channel our energy and resources behind ambitious greenfield efforts like these.

For Incubation track, we are looking for visionary ideas backed by driven, mission-aligned people able to think outside of the status quo. Applicants may apply individually or as a founding team. Strong candidates will have an established track record of safety-relevant research, high agency, and an original perspective (additional co-founders in non-research roles are welcome). Great agendas will mitigate important risks in novel or neglected ways. We are thrilled to support residents drive towards founding a new organization for their program, but this is not necessarily required: the organizing motivation is to open new branches of the tech tree with creative, foundational work.

This track maintains a high bar. We aim to ensure that the most promising and feasible ideas are championed. Residents will do exploratory and proof of concept work, meet widely with top experts to sharpen their approach, disseminate frameworks for thinking about the problem alongside any empirical results, and work with our team to aggressively scale scope and ambition of the primary research path. All Incubation Residents submit a Governance and Publication Plan at admission. This covers oversight arrangements, publication policy, and the mitigation strategy for information hazards and dual-use research risks.

• Duration · 6 to 24 months
• Review Cycle · 12 weeks
• Key feature · Funding Evaluation Workshop at end of Block 1
• Required at admission · Governance and Publication Plan, Founding Rationale, 90-day plan with explicit stop/pivot conditions