Research institute · Research orgs · London, United Kingdom · Est. 2013
The Center on Long-Term Risk (CLR) is a UK-registered charity based in London that conducts interdisciplinary research and grantmaking focused on worst-case, long-term risks from advanced artificial intelligence—often framed as “s-risks” (risks of astronomical suffering) and, operationally, as pathways to reduce catastrophic AI conflict and related failures. CLR’s work combines (1) research agendas aimed at identifying and steering model behavior in high-stakes settings and (2) a pipeline of community-building and capacity-building programs designed to grow expertise in AI risk reduction.
As of 2026, CLR describes its research as running across two streams: an empirical “Model Personas agenda,” which studies how models can develop undesirable behavioral traits/personas during training, and a conceptual “Safe Pareto Improvements (SPIs) agenda,” which develops game-theoretic and decision-theoretic interventions intended to make cooperation and “safe” bargaining more robust in advanced AI interactions. In addition to research, CLR operates grantmaking through the “CLR Fund,” and runs programs such as a Summer Research Fellowship, a Research Affiliates program, and reading/fundamentals programs (including an SPI-focused fundamentals course). CLR positions itself within the broader AI safety and longtermist ecosystem primarily by aiming to be influential in technically specific, externally legible areas—especially those it argues are relevant to preventing catastrophic conflict between powerful AI systems.
No curated programs yet. Check the careers page ↗.
Why might an agent expect opting out of safe Pareto improvements to have better consequences?
Don’t Inoculate Everything: Stratified Inoculation Prompting Narrows Backdoors and Preserves Desired Traits — Center on Long-Term RiskInoculation prompting applies one prompt to every training example, which leaves a backdoor through which similar prompts still elicit the undesired trait, and…
Taboo “equilibrium”: Less confused frames for research on AI bargaining — Center on Long-Term RiskCommon frames on bargaining problems — especially the assumption that agents play an equilibrium — are confused, and get in the way of research on safe Pareto…
Summer Update 2026 — Center on Long-Term RiskWe're writing to share our research progress and program updates from the first half of 2026.
Value Leakage: An LLM's Answers Are Silently Shaped by Its Own Values — Center on Long-Term RiskPeople use language models for practical questions whose answers are difficult to verify.
Inoculation Adapters: Improved Selective Generalization of Capabilities with Fewer Surprising Backdoors — Center on Long-Term RiskInoculation prompting is a selective-generalization technique used against Emergent Misalignment.
A high-level model of AI bargaining — Center on Long-Term RiskAdvanced AIs might be capable of various credible commitments unavailable to humans, which they could use when bargaining with each other.
Evals for “SPI-incompatible” behavior and reasoning — Center on Long-Term RiskIn Part I of CLR's safe Pareto improvements (SPI) agenda, we gave our high-level strategy for evaluating models for SPI-incompatible behavior and reasoning.
Strategic Obfuscation of Deceptive Reasoning in Language Models — Center on Long-Term RiskLarge language models can exhibit different behaviors during training versus deployment, a phenomenon known as alignment faking.
Safe Pareto Improvements Research Agenda — Center on Long-Term RiskSafe Pareto improvements (SPIs) are modifications to agents’ bargaining strategies that make all parties better off, regardless of their original strategies.
Shaping the exploration of the motivation-space matters for AI safety — Center on Long-Term RiskWe argue that shaping RL exploration, and especially the exploration of the motivation-space, is understudied in AI safety and could be influential in…
Model Persona Research Agenda — Center on Long-Term RiskCLR’s overall mission is to reduce the risk of astronomical suffering from powerful AI, or s-risks.
CLR published a mid-2026 update describing research progress across two streams, noting a research team change (Daniel Tan on sabbatical; Vili Kohonen joining in June) and describing program activities including a SPI Fundamentals Program run in August and the launch of a Research Affiliates program.
Annual Review & Fundraiser 2025: staff transition; research agenda clarification; fundraising target for 2026 budgetCLR’s 2025 annual review and fundraiser describes transitions in leadership (Jesse Clifton stepping down as Executive Director; Tristan Cook and Mia Taylor taking over roles; later Mia’s departure), summarizes research agendas (AI personas/emergent misalignment; safe Pareto improvements), and states a 2026 fundraising aim of $400,000 to achieve its target budget for 2026.
Summer Update 2025: Mia Taylor departure announced; Tristan Cook to lead after transitionCLR reports that Mia Taylor decided to leave at the end of August, joins Forethought, and states that Tristan Cook will lead after the transition; the post also summarizes empirical personas and conceptual strategic readiness work during her interim research director period.
Center on Long-Term Risk: 2025 Plans sets focus on externally legible empirical agenda and ‘strategic readiness’CLR outlines plans for 2025, including building expertise and external collaborations via an externally legible empirical agenda (LLMs/personas and related directions) and continuing a strategic readiness agenda focused on when interventions robustly reduce s-risk.