Founding Member, Technical Staff

Adapted from the reading menu in gsarti/gsarti.github.io.

Location: Remote, Europe/UK (London, UK in-person preferred)
Salary: $100,000 – $135,000 gross annual base salary, depending on experience and location
Contract: Full-time, initially for two years; extension possible subject to continued funding
Start date: As soon as possible, ideally by November 2026

Apply by September 27th, 2026

For questions about this position, please contact hr@parallx.ai.

About Parallax

Parallax is a new UK-based non-profit research organisation developing open methods and infrastructure for white-box auditing of frontier AI systems. Our mission is to make evidence about models’ internal beliefs, goals, and plans a core component of AI evaluations, allowing evaluators to identify, explain, and predict safety-relevant failures more reliably.

We publish our findings, release our methods and infrastructure, and aim to make scientific contributions that remain useful as agentic capabilities of frontier models keep increasing. You can read more about our mission, strategy, and team on our About page.

About the role

We are recruiting two Members of Technical Staff to join Parallax as founding members of our technical team. As an early member of a fast-moving, highly collaborative team, you will take substantial ownership of projects developing frontier white-box AI auditing technology, from ideation to implementation, evaluation and release. You will also play a fundamental role in defining Parallax’s character and impact as it scales into a leading European actor in this space.

You will work directly with our co-founders: Mario Giulianelli, our Scientific Director, is an Associate Professor at UCL who was previously a Senior Researcher at the UK AI Security Institute (UK AISI). Gabriele Sarti, our Technical Director, was a postdoctoral researcher at Northeastern University and the National Deep Inference Fabric (NDIF).

Our advisors are David Bau, Yonatan Belinkov, Seraphina Goldfarb-Tarrant, and Cozmin Ududec. Our existing collaborations include the National Deep Inference Fabric (NDIF) at Northeastern University and Cohere’s AI Safety team. We also plan to engage with organisations across the evaluation and interpretability ecosystem, including the UK AI Security Institute, Meridian Labs, Goodfire, and METR.

About our research

You will lead research efforts to elicit and test high-level latent variables to explain model behaviour, including beliefs about the current and future action-conditioned environment states, plans, goals, and motivational structures. Examples of our current priorities include:

• Improving and validating meta-models such as activation oracles to decode model representations into concise, testable descriptions.
• Developing counterfactual training methods to improve meta-models’ predictivity, robustness and causal sensitivity.
• Creating benchmarks to validate whether elicited beliefs reliably predict model behaviours, distinguishing between competing explanations for different failure categories.
• Developing scalable methods for locating pivotal steps for efficient belief elicitation in long-horizon agent trajectories.
• Building open research infrastructure to make proposed methods reproducible and usable by other evaluators.

As a founding technical team member, you will help refine these questions and shape Parallax’s broader research agenda.

Responsibilities

Your main responsibility will be to proactively identify and execute impactful research alongside the Parallax team. This includes:

• Taking ownership of substantial research lines, from scoping and prototyping through experimental validation and release.
• Collaborating with founders to formulate clear scientific questions and design rigorous experimental protocols to validate them.
• Designing and implementing efficient and reliable training and inference pipelines, supporting fast iteration while preserving legibility and reusability.
• Running large-scale experiments using inference APIs and multi-node GPU infrastructure.
• Contributing to write-ups, benchmarks and open-source software to share the results of your analyses, and presenting your research at conferences and research convenings.
• Helping supervise research interns, partnering with external collaborators and contributing to the development of future members of the technical staff.
• Participating in decisions about Parallax’s research direction, technical standards, and organisational development.

• A strong sense of ownership, initiative, and reliability, with an honest desire to make a concrete impact on pressing issues.
• A commitment to research rigour and intellectual honesty, with a tendency for meticulous experimentation, holding themselves to a high bar for final results and write-ups.
• Strong written and verbal communication, low ego, and openness to feedback.
• Ability to learn quickly from teammates and mentors, and mindfully adopt new technologies to accelerate work.
• The ability to move rapidly from scrappy prototyping to careful implementation and robust experimentation, keeping the bigger picture in mind when refining low-level details.
• Deep domain knowledge and research experience in areas such as interpretability, reinforcement learning and agent evaluations.
• Strong programming ability (in particular Python and PyTorch/HF Transformers ecosystems), with the ability to navigate efficiency and software design trade-offs.
• Experience designing and running large-scale machine learning experiments using multi-GPU setups and making efficient use of inference APIs.
• The ability to identify useful research questions and make progress when the path is unclear or blocked.
• A proven interest in collaborative research, a commitment to openness and the willingness to contribute across scientific and engineering work.

A PhD in machine learning or a related field is not required. We welcome applicants who can demonstrate equivalent research ability through other professional routes. If the role interests you but you do not meet every criterion, we still encourage you to apply.

Location and working arrangements

We prefer candidates who can work regularly from the London AI Safety Initiative (LISA) offices. We are also open to remote candidates based elsewhere in Europe.

Remote team members will be expected to attend an initial team gathering in London and visit London regularly for collaborative research, planning, and team activities. Parallax can sponsor UK visas through Meridian Impact CIC and, where appropriate, employ international candidates through an employer of record.

We offer flexible working hours and place greater emphasis on reliable collaboration and high-quality results than on a uniform daily schedule, but candidates must be able to work regularly with colleagues in European time zones.

Compensation and benefits

The gross annual base salary for this position ranges between $100,000 – $135,000, depending on experience and location.

The role also includes:

• A 3% annual salary increase.
• A 5% end-of-year bonus.
• A 3% employer pension contribution.
• Unlimited PTO.
• £1,000 yearly personal development budget.
• A capable work laptop for accelerating your research.
• Substantial access to compute and API resources.
• Support for conference attendance, research travel, and publication.
• Covered lunch, snacks and coworking from the LISA offices if in person.
• Visa sponsorship or international employment support where applicable.

Hiring process

Our expected hiring process is:

1. An initial review based on your application.
2. An introductory interview focused on your motivation and relevant experience.
3. One paid take-home exercise of 4 hours, reflecting the research and technical work of the role.
4. A research discussion of your approach, results, and possible next steps
5. A second paid take-home exercise of 2 days.
6. A final interview to discuss your results and explore your fit with the team.

Exercises are remote and compensated with £180 and £540, respectively. We will agree on a completion window that fits around your existing commitments.

Equal opportunities

Parallax is committed to building an inclusive working environment. We welcome applications regardless of race, ethnicity, nationality, gender, gender identity, sexual orientation, disability, religion, age, socioeconomic background, or educational pathway.

Where possible, we will make reasonable accommodations throughout the application and interview process.

Apply by September 27th, 2026