RocketHuntSmart job search
Principles of Intelligence (PrincInt)
5 days ago
6,250 – 9,500 USD
per month
6250$-9500$/month remote PIBBSS recruiter contact Horizon Scanner (Contractor)
Principles of Intelligence (PrincInt) is hiring a small cohort of researchers to conduct sensemaking and horizon scanning on interdisciplinary approaches to scalable alignment and related areas during the pilot phase of four (4) new Coordinated Research Programs (CRPs).
Over a paid ~four (4)-month engagement, you will build a deep and complete understanding of the bets the field is making in one of our new research areas, lay out a roadmap for how the technical work should be done, and find the expertise communities outside AI safety that should be working on it.
If successful in this role, you may be offered an extension and promotion to Technical Director of the Coordinated Research Program you helped scope. As Technical Director, you would lead your own research team and research agenda within PrincInt and steward research groups incubated through the program.
Location: Remote, with no location requirement.
Travel: You’ll be expected to attend at least one convening or horizon-scanning event in person during the engagement. Travel and accommodation are covered.
Type: ~Four (4)-month engagement. Contractor. Full-time preferred; part-time considered for strong candidates with existing academic or research work.
Start date: Negotiable, with a strong preference for starting as soon as possible. For a sense of timing, an engagement that begins in mid to late November 2026 would run until around mid to late March 2027.
Compensation: $25,000–$38,000 for ~four (4) months, depending on seniority and cost of living in your location, plus travel costs. Exceptional candidates may earn more.
Extension and promotion: The contract may be extended depending on fit for both sides, with potential promotion to Technical Director of the Coordinated Research Program you helped scope. Technical Director salaries sit above the rate for this engagement.
Reporting line/collaborators: Reports to the Executive Director, Research, Lucas Teixeira. You’ll work alongside Lauren Greenspan (Technical Director), Cecilia Wood (Technical Director, Mechanism Design), and Daniel Herrmann (Horizon Scanner, Philosophy).
Application deadline: Please submit application by 5 November 2026. Applications will be reviewed on a rolling basis. Earlier applications have an advantage.
Referrals: We’re happy to receive referrals.
Principles of Intelligence (PrincInt) is a 501(c)(3) nonprofit research organization that aims to grow the AI safety field by fostering interdisciplinary engagement with the natural and social sciences. We support foundational scientific contributions that aid the development of safe and globally beneficial AI systems. Through our research programs, residencies and field-building efforts, we bring together expertise from different disciplines to advance understanding of intelligent behavior and strengthen the scientific foundations required to create aligned AI systems. We began in 2021 as PIBBSS, and took the name Principles of Intelligence as the work expanded beyond the fellowship into coordinated research programs, an internal research division, the incubation of new research organizations, and fiscal sponsorship for independent labs and researchers.
We develop and support coordinated research programs (CRPs) which are built around the principle of use-inspired basic research: fundamental scientific exploration in service of creating AI safety-relevant public goods (e.g., high-assurance interpretability tools, robust alignment targets). This work aligns with our PIBBSS fellowship and expands on past field-building efforts, which focused on supporting individuals rather than groups.
Developing a CRP runs through three phases:
A cohort of Horizon Scanners begins scoping the program, defining what success would look like and identifying relevant expertise communities, before reaching out through discipline-specific conferences, workshops, and one-on-one conversations with potential collaborators. Seed funding, in turn, supports an initial internal research team that gives the program a base of expertise. By the end of this stage, we aim to show external collaborators that the program is credible and its problems tractable.
Once the program has a shared vision and interest from several research groups, Horizon Scanners work closely with expert collaborators to develop a detailed roadmap that sets out what is known, where the gaps lie, and which approaches to pursue. The program then assembles a portfolio of scoped research projects led by different Principal Investigators (PIs), each addressing the program’s core bottlenecks from a complementary angle, while semi-private workshops and public satellite events bring more of the community into the work.
The program has reached carrying capacity: A stable portfolio of funded research projects, a connected community of collaborators, and a clear theory of how the outputs will reach their intended users. Horizon Scanners shift emphasis toward evaluation and transition: tracking which research threads are delivering and which are not, and beginning substantive conversations with the labs and regulators who will need to absorb the results.
Our first research program, Physics-Informed Ambitious Mechanistic Interpretability (PIAMI), has recently entered the sustaining phase and applies methods from statistical physics, supports work from university groups, collaborates with the AI safety community, and includes an internal full-time research division (PIRAMID).
We are now looking to expand to four (4) new research programs, with the goal of launching RFPs and incubating new groups by the next academic hiring cycle.
All four are open for this cohort.
What does it mean for an AI system to be aligned, and what should we align it to? We are particularly interested in alignment targets that remain stable under continual learning and leave a verifiable fingerprint in a model’s internals.
Example frameworks: CEV, corrigibility, Constitutional AI, reward function design.
Example target communities: Social choice theory, ethics, political philosophy, neuroscience, law.
What is an agent? When do our measurements and evaluations tell us something ‘real’ about models’ intents, beliefs, desires, etc.? How do such properties emerge from lower-level dynamics? How can we extend our understanding of agency so that our safety properties remain robust as systems change through an intelligence explosion?
Example frameworks: Radical interpretation, decision theory, agent foundations, embedded agency, AIXI, causal models, world models.
Example target communities: Philosophy (decision theory, formal epistemology), artificial life, theoretical reinforcement learning, biology, complex systems.
How do we gain reliable epistemic access to systems we cannot fully observe? When do our evaluations and interpretability measurements actually license the safety claims made on their basis? What can the history of science teach us about how progress is made on problems like alignment, and which epistemic strategies best suit the field?
Example target communities: History and philosophy of science, science and technology studies.
How must our economic, legal, and political institutions adapt once cognition is no longer a bottleneck? How do we preserve meaningful human agency and guard against gradual disempowerment?
Example target communities: Macroeconomics, law, sociology, anthropology, political theory, science and technology studies.
This is the pilot phase, and you are the person doing it.
Each Horizon Scanner takes on one of the aforementioned programs, mapping its scientific landscape and the safety case it should serve, setting out how the technical work should proceed, and identifying both the groups already working on these problems and the expertise communities outside AI safety whose methods the program could draw on.
We run events and outreach alongside this work, and if the program succeeds, you may become its Technical Director.
Sensemaking and opportunity mapping. You will map existing work in your program area, locating its technical bottlenecks and assessing which problems are neglected and tractable, which, in turn, allows us to define the research outputs that would constitute progress.
Horizon scanning and network identification. You will identify the researchers, groups, and adjacent fields relevant to the program and assess what each can contribute, thus determining where we should collaborate, whom we should recruit, and where targeted support is needed.
Agenda curation and technical translation. You will state the alignment problem in terms that researchers from other disciplines can work with, clarifying the definitions, assumptions, threat models, and success criteria involved, as well as their bearing on the AI systems being built today.
Engaging the network. You will convene prospective participants, publish a roadmap or request for proposals where appropriate, and support researchers in developing their proposals, agreeing with them in advance on the technical scope of the work and on how we will evaluate it.
Over four months, we expect some subset of the following, scoped with us at the start:
Plus
We also expect you to run an event, independently or with us, and to compile a list of the people and groups worth funding, along with your assessment of each.
As Technical Director, you would lead your own internal research team as part of a research division within PrincInt, and continue to liaise with the research groups incubated through the program. Alongside this, the role also involves:
This horizon scanning contractor engagement serves as a trial for the Technical Director role, a full-time core team position.
We care most about whether you can build a real map of a problem space, and then get the right people to care about it. We expect strong candidates to have many of the following:
Compensation for the ~four (4)-month engagement is $25,000–$38,000, depending on prior experience and location, plus travel costs. We cover travel and accommodation for events on top of that, and we can support short-stay visa arrangements where required.
Flexible working hours: Flexible working hours with an emphasis on outcomes and reliable coordination with teammates.
High ownership & autonomy: A high-trust environment where you’ll own your work end-to-end, with meaningful responsibility and room to improve systems.
Team meetups: We get together in person 2–4 times a year for focused work time, great conversations, and the kind of momentum you only get when the whole team is in the same room.
On conversion to Technical Director: 4 weeks of collective holiday per year plus additional personal time off, and the rest of our core team benefits.
The application form takes 45–60 minutes.
Application. We review on a rolling basis until 5 November 2026.
Interview. Successful applicants are invited to a remote interview.
Work trial. A paid, one-day remote work trial: a scoped piece of the actual sensemaking work in your area, plus time with the team.
Offer.
References: We’ll ask for two or three in the form, and will let you know before we contact them. Good references can notably strengthen your application.
PrincInt was founded as PIBBSS in 2021 by Nora Ammann, Tushita Jha (TJ) and Anna Gajdova as a research initiative aiming to draw insights & talent from fields studying intelligent behavior in natural systems towards progress on questions in AI risk and safety. Since then, we have raised more than 5.5M USD, supported ~70 researchers for a three (3)-month, full-time PIBBSS fellowship, first run in 2022, and organized 20+ AI safety research events/workshops. Over the years, we have built a substantive research community, spanning many disciplines, across academia and industry, both inside and outside of AI safety.
Lucas Teixeira (Executive Director, Research) Lucas has led PrincInt’s research strategy since 2023. They ran the research affiliateship that grew into PIRAMID, launched PIAMI, and now act as a research manager and collaborator across PrincInt’s research threads, drawing on the history and philosophy of science to support pluralistic and non-paradigmatic research practices. Before joining PrincInt, they worked at Conjecture, first as an Applied Epistemologist, where they helped co-run Refine, and later as a Research Engineer on the Cognitive Emulation agenda. They studied philosophy, anthropology, computer science and political science before devoting themselves to the alignment problem.
Lauren Greenspan (Technical Director) Lauren brings a varied perspective to AI safety research with her background spanning physics, machine learning, and science, technology, and society (STS). She is particularly interested in physics approaches for interpretability, the evolution of scientific
RocketHunt estimate from real salaries
Work Format
Remote
Employment
Full-time
English Level
B2 - Upper-Intermediate
AI writes it from your resume. You only have to send it.
