Bridging the Gap: AI Scenario based Policy Pipeline

Bridging the Gap is a proposed project that creates a pipeline of rigorous, near-term forecasts of catastrophic AI risk that are then used to create decision-ready policy advice for government policymakers. Over 12 months, the Swift Centre's professional forecasting team runs continuous 8–12-weekly cycles, each pairing a set of bespoke AI-risk forecasts with an open policy competition. The project would
provide an accessible, low-barrier pathway for participants to develop practical and robust AI policy proposals, alongside creating a bank of unique options and analysis for policymakers to draw on.

This expands an already successfully pilot of this project earlier in the year where we produced 5 bespoke forecasts, undertook targeted outreach, received 29 individual policy advice submissions, and assembled a judging panel of experienced professionals across national security, military, and AI policy to evaluate them.

Scenarios will be chosen in consultation with our existing network of experts and contacts working on AI policy inside central governments, non-profits, think tanks, and our own team. The Swift Centre's forecasting team then will produce detailed, rigorous forecasts on the likelihood of these scenarios and the explicit assumptions behind them. Participants learn optimal policymaking skills by drawing on these scenarios, the forecasts, rationales, and their wider research/knowledge to create 3-page policy briefs tailored to specific decision-makers (e.g. a defense secretary or an AI minister). The strongest submissions are connected directly to government stakeholders (e.g. UK AISI, DSIT, US Department of War) through closed-door roundtables and networking events.

The project will be led by Executive Director James Newport, drawing on the Swift Centre team (Founder/Director Michael Story; COO Eleanor Parr) and its established forecasting infrastructure.

James Newport - Project Lead (Executive Director, Swift Centre). Leads the Swift Centre's integration of scenario analysis and forecasting into decision-making for governments, Google DeepMind, and major AI-safety philanthropic funders. He has provided advice to a core UK public sector institution on how to integrate the best scenario analysis and foresight approaches into their decision/policy making practices. Previously Head of Policy at ControlAI (co-author of A Narrow Path); Head of Development Policy at HM Treasury, leading the UK's £3.5bn financial response to the war in Ukraine and managing the UK's 0.7/0.5% international aid budget; Head of Strategy at the Centre for Data Ethics and Innovation. MSc Cognitive and Decision Sciences (UCL, Distinction, Dean's List), with dissertation research on improving Civil Service forecasting accuracy via Bayesian reasoning.

The Swift Centre forecasters include:

• Lead financial institutions and policy analyst at the US Federal Reserve Board; 

• Former US Navy nuclear submarine commander; 

• Senior Scientist (Emeritus), University of Wisconsin–Madison (biochemistry, virology, immunology); 

• PhD European lawyer and an AI trends researcher for a leading AI evaluator

• Neuroscience PhD and former US affiliate scientist on influenza at Los Alamos National Laboratory; 

• PhD philologist/linguist (cultural bias in forecasting) and RAND policy analyst; 

• AI policy researcher and forecaster for AI 2027 (MSc computational and applied mathematics); 

• Former AI & Digital Transformation Lead and Global Risk & Scenario Forecaster at the Forecasting Research Institute.

The concrete outputs over the project would be:

1. A recurring stream of publicly available forecasts on AI-risks/scenarios;
2. A growing bank of decision-ready policy briefs government civil servants and politicians can easily access and use - due to both the Swift Centre network and roundtable events, and also because the policy templates we will provide participants will align with actual real-world advice templates and process;
3. A low-barrier, testable route for new and existing talent to prove their fit for AI policy and governance work, with potential to agree tie-ins with talent pipelines (BlueDot Impact, Talos Fellowship, GovAI).

Preventing catastrophic and existential risks from transformative AI (e.g. severe biosecurity threats, autonomous cyber-attacks, loss of control, power concentration etc.) requires policy that is actionable, resilient to rapid capability shifts, and grounded in robust, explicit probabilistic reasoning and not vague warnings.

All policy rests on two predictions: what the future will be, and how an intervention will change it. Our scenario based forecasts supply the first in a robust, trusted, and transparent way; while participants help to crowd source options by writing advice on what actions can be taken to change or react to said future. The resulting briefs are directly usable by decision-makers as they are tailored to align with the exact format and approach that policy officials will provide to Ministers/Secretaries of State, whilst also ensuring the advice itself is explicit in the assumptions each option is built on and how it directly influences the risk that has been identified.

The project reduces catastrophic and x-risk through two channels.

1. Through capacity building: it gives non-technical professionals a frictionless way to test and develop the rare skill of writing concise, well-reasoned AI policy advice. It builds an on and off ramp from existing channels that build talent in the AI safety field (e.g. BlueDot, Talos, GovAI etc.) and provides hiring organisations with a public pool of people who they can verify are able to design tangible policy solutions.
2. Through direct scenario analysis and policy options:** each cycle generates a bank forecasts on important AI risk scenarios which are independently useful for decision makers, whilst also building upon this by facilitating the creation of clear policy options. Through the Swift Centre network, we can help deliver the best policy briefs to officials in the UK and US governments, improving the quality and speed of the governance response as capabilities advance.

Full budget: https://docs.google.com/spreadsheets/d/1VV04Km2AGRMI781wWf_jRCuJqHBumY3p4x2H3c-RLYs/edit?usp=sharing

Minimum ($80,000 / ≈£60k) would allow us to run the core at the previous scale. Covers operations, overheads, and roughly two full forecasting-plus-policy cycles that would include 5 questions each. Each scenario/forecasting round costs ≈£25,000, covering the commissioned Swift Centre professional forecasters and basic prize incentives. This funds the open platform with the bespoke AI-risk scenarios and forecasts, full rationales, and the policy submissions.

Ideal ($201,000 / ≈£150k) would allow us to scale to continuous cycles plus deeper integration with decision makers. Everything in the minimum but we'd run up 7 scenario based forecasting rounds (each with 5 questions) over 8–12 weekly cycles across a 12 month period. In addition, this additional funding would allow us to: (1) explore formal partnerships with talent organisations (BlueDot Impact, Talos Fellowship, GovAI - all of whom we have existing relationships with) so the project could serve as a fit-test, an applicant-assessment tool, and a follow on project for their course attendees/fellows; and (2) run monthly bridge-building events and closed-door roundtables connecting the best submissions directly to government decision-makers (e.g. UK AISI, DSIT) and organisations hiring in the space.