Sentinel (Catastrophic Risk Foresight)

Quick Assessment

Key Links

Overview

Sentinel is a foresight and emergency response organization focused on anticipating and mitigating large-scale catastrophes, with particular attention to tail-risk scenarios.¹ Co-founded by Nuño SemperePersonNuño SempereNuño Sempere is a Spanish superforecaster who co-founded the highly successful Samotsvety forecasting group and now runs Sentinel for global catastrophe early warning, while being known for skeptic...Quality: 50/100 and Rai Sur, the organization began operating informally around 2023 and incorporated as a US 501(c)(3) nonprofit in or after late 2024.²³ Sentinel combines automated intelligence gathering with human probabilistic judgment: language models process large volumes of news items weekly for initial triage, and the most credible signals are then routed to forecasters who assign calibrated probabilities to emerging threats.¹

Sentinel was founded by Sempere, a co-founder of Samotsvety ForecastingOrganizationSamotsvetyElite forecasting group Samotsvety dominated INFER competitions 2020-2022 with relative Brier scores twice as good as competitors, providing influential probabilistic forecasts including 28% TAI by...Quality: 61/100 — a superforecasting group that won the CSET-Foretell forecasting competition in 2020 and 2021, described by Scott Alexander as winning "by an absolutely obscene margin, around twice as good as the next-best team."⁴ Sentinel carries forward that group's emphasis on quantified risk assessment. Its core architecture has two stages: high-throughput machine scanning for weak signals, followed by deliberative probabilistic reasoning from domain-aware forecasters. This separates Sentinel from both purely algorithmic monitoring services and conventional research organizations, neither of which routinely pairs broad surveillance with calibrated probability estimates on tail risks. The intended output is probability estimates expressed as confidence intervals rather than qualitative assessments.¹

As of November 2025, Sentinel reports approximately $700K on its balance sheet against a full funding target of approximately $1.6M, and its newsletter has grown from 552 to 3,936 subscribers since its first major public fundraise in November 2024.⁵

Organizational Structure and Funding

Sentinel is a registered US 501(c)(3) nonprofit organization, incorporated in or after late 2024.³ The incorporation followed a period of informal operations bootstrapped by Nuño SemperePersonNuño SempereNuño Sempere is a Spanish superforecaster who co-founded the highly successful Samotsvety forecasting group and now runs Sentinel for global catastrophe early warning, while being known for skeptic...Quality: 50/100's consulting income and early crowdfunding.⁶ The organization accepts donations via wire transfer and Manifund, an effective altruism-aligned funding platform.³

Sentinel's budget is organized around four verticals: Sustainable Leadership, AI OSINT, Foresight Team, and Content & Distribution.⁵ As of November 2025, the organization had raised approximately $700K of a $1.6M full-funding target.⁵ Primary funding has come from individual donors and EA-adjacent crowdfunding; no grants from large institutional funders such as Coefficient GivingOrganizationCoefficient GivingCoefficient Giving (formerly Open Philanthropy) has directed \$4B+ in grants since 2014, including \$336M to AI safety (~60% of external funding). The organization spent ~\$50M on AI safety in 2024...Quality: 55/100 have been publicly disclosed.

The organization is registered in New York (per its @XriskFYI X/Twitter account, which describes itself as a "Media & News Company based in New York City").⁷ The team is geographically distributed: co-founder Rai Sur is based in New York,⁸ while forecaster Vidur KapurPersonVidur KapurVidur Kapur is a superforecaster and AI policy researcher involved in multiple forecasting organizations and the Sentinel early warning system, contributing to AI risk assessment and EA Forum discu...Quality: 38/100 is based in the United Kingdom.⁹

Sentinel has two structural components:¹⁰

• Foresight Team: Monitors current events and produces weekly risk assessments. Alerts the emergency response team when assessed risk crosses defined thresholds.
• Emergency Response Team: A reserve team with diverse skill sets intended to search for high-impact actions during crises. Described as "somewhat inactive" on the official website as of early 2025, pending further development.³

Core Activities

Risk Detection Pipeline

Sentinel operates a structured, multi-stage pipeline for identifying and assessing global catastrophic risks.¹ Automated ingestion draws on millions of news items, academic preprints, policy documents, and social media signals each week across multiple languages and regions. Large Language ModelsConceptLarge Language ModelsComprehensive assessment of LLM capabilities showing training costs growing 2.4x/year (\$78-191M for frontier models, though DeepSeek achieved near-parity at \$6M), o3 reaching 91.6% on AIME and 87...Quality: 62/100 filter this volume down to a prioritized shortlist, flagging items against predefined criteria — scale, novelty, irreversibility, and escalation potential. Forecasters with demonstrated calibration track records then evaluate each flagged item, assign probability estimates grounded in reference-class reasoning, and situate developments within longer-term risk trajectories. Final outputs are packaged into structured briefs distributed on a weekly cadence.¹

1. Automated Collection: Millions of news items, preprints, and policy documents ingested weekly across multiple languages and source types.
2. AI Prioritization: Language models filter and rank candidate events against standing catastrophic-risk criteria, including scale, novelty, and escalation potential.
3. Forecaster Assessment: Human forecasters evaluate flagged items, assign calibrated probability estimates, and update trend lines for tracked scenarios.
4. Publication: Curated weekly summaries distributed via newsletter and podcast.

In addition to the standard pipeline, Sentinel has been developing an AI-enabled wargaming tool called grim to scale up the number of catastrophic scenarios that can be explored and to improve emergency response capabilities.¹¹

Weekly Briefs

Each weekly brief is organized around a consistent set of risk categories, enabling readers to track developments across domains without re-orienting each issue. As of 2024–2025, Sentinel has tracked threats including H5N1, solar flares, militant terrorism, AI incidents and governance developments, and indicators of potential Chinese invasion of Taiwan.¹²

Central to each brief is the risk status indicator — a color-coded ordinal signal consolidating the forecasting team's aggregate view of near-term catastrophic risk into a single at-a-glance readout. The indicator surfaces directional changes week over week, functioning as a running thermometer for civilizational-scale threat levels. Significant movements are accompanied by brief explanatory notes attributing shifts to specific developments in the underlying risk categories.

Track Record

No publicly available formal track record report, Brier scores, or calibration metrics have been published by Sentinel as of this writing. Sentinel's November 2024 funding memo states that "judgemental forecasting by Sentinel's expert forecasters has a very good track record for short-term forecasts," but does not provide quantitative supporting data.¹⁰ One documented predictive success cited by Sentinel is correctly anticipating the WHO's decision to rate monkeypox a Public Health Emergency of International Concern (PHEIC) in 2024.¹²

Team

Sentinel's team combines probabilistic forecasting, AI policy research, and technology backgrounds. Forecasters appear to be drawn primarily from Samotsvety and similar superforecasting circles, which recruit based on demonstrated track records on platforms including Good Judgment Open, INFER, and MetaculusOrganizationMetaculusMetaculus is a reputation-based forecasting platform with 1M+ predictions showing AGI probability at 25% by 2027 and 50% by 2031 (down from 50 years away in 2020). Analysis finds good short-term ca...Quality: 50/100.¹³

• Nuño Sempere — Co-founder and Head of Foresight. Co-founded Samotsvety ForecastingOrganizationSamotsvetyElite forecasting group Samotsvety dominated INFER competitions 2020-2022 with relative Brier scores twice as good as competitors, providing influential probabilistic forecasts including 28% TAI by...Quality: 61/100 with Misha Yagudin, having met at a 2020 summer fellowship at Oxford's Future of Humanity InstituteOrganizationFuture of Humanity InstituteThe Future of Humanity Institute (2005-2024) was a pioneering Oxford research center that founded existential risk studies and AI alignment research, growing from 3 to ~50 researchers and receiving...Quality: 51/100. Previously a researcher at the Quantified Uncertainty Research Institute (QURI), where he built Metaforecast.org, a search tool aggregating predictions from multiple forecasting platforms. Bootstrapped Sentinel through his consulting practice, Shapley Maximizers, before going full time on Sentinel in mid-2024.⁶¹³
• Rai Sur — Co-founder. Previously CTO of Alongside (later renamed Universal), a crypto index platform that raised $11M led by a16z crypto in 2023. Educated at The Cooper Union for the Advancement of Science and Art. Joined Sentinel as co-founder in mid-2024, initially committing to three months full-time.⁸¹⁴
• Vidur KapurPersonVidur KapurVidur Kapur is a superforecaster and AI policy researcher involved in multiple forecasting organizations and the Sentinel early warning system, contributing to AI risk assessment and EA Forum discu...Quality: 38/100 — Foresight team member. Forecasts for Good Judgment, Swift CentreOrganizationSwift CentreSwift Centre is a UK forecasting organization that provides conditional forecasting services to various clients including some AI companies, but is not primarily focused on AI safety. While they de...Quality: 50/100, SamotsvetyOrganizationSamotsvetyElite forecasting group Samotsvety dominated INFER competitions 2020-2022 with relative Brier scores twice as good as competitors, providing influential probabilistic forecasts including 28% TAI by...Quality: 61/100, Sentinel, and the RAND Forecasting Initiative, as well as a hedge fund. Studied medicine and public health. Based in the United Kingdom.⁹
• Tolga Bilge — Foresight team member. AI Policy Researcher at ControlAIOrganizationControlAIControlAI is a UK-based advocacy organization that has achieved notable policy engagement success (briefing 150+ lawmakers, securing support from 100+ UK parliamentarians) while promoting direct in...Quality: 63/100. Also a professional forecaster for the Swift CentreOrganizationSwift CentreSwift Centre is a UK forecasting organization that provides conditional forecasting services to various clients including some AI companies, but is not primarily focused on AI safety. While they de...Quality: 50/100, the RAND Forecasting Initiative (formerly INFER), and Good Judgment Inc. (Superforecaster). Holds a Bachelor's in Mathematics from the University of St Andrews and a Master's in Mathematics from the University of Bergen. Main research interests include Technical AI Safety and governance and existential risk reduction.¹⁵
• belikewater — Foresight team member; a pseudonym used by an anonymous forecaster previously affiliated with SamotsvetyOrganizationSamotsvetyElite forecasting group Samotsvety dominated INFER competitions 2020-2022 with relative Brier scores twice as good as competitors, providing influential probabilistic forecasts including 28% TAI by...Quality: 61/100. Described as "an anonymous forecaster known as belikewater" in Nuño Sempere's July 2022 EA Forum forecasting newsletter, where the forecaster contributed analysis on monkeypox spread. No real name has been publicly disclosed.¹⁶ A podcast transcript reveals "belikewater" to be the forecasting moniker of a foresight team member named Lisa.¹⁷
• Leif Sigrúnsson — Foresight team member (listed on official website).¹³

Podcast episodes are collaboratively written with contributions from a wider group including Brian Yu, Kati Conen, Alex Lyzhov, Jaeho Lee, and 30+ others.¹⁸

Media and Distribution

Newsletter

Sentinel publishes via Substack at blog.sentinel-team.org. As of November 2025, the newsletter had approximately 3,936 subscribers, up from 552 in November 2024 — a 7.1x increase.⁵ As of November 2024, the newsletter had no paid/paywalled tier; Sentinel indicated it planned to introduce a pledge mechanism pending bank account setup.¹⁹ Donations are accepted via Manifund and wire transfer.³

Sentinel Minutes Podcast

Sentinel Minutes is a podcast available on Apple Podcasts (ID: 1808187161), Spotify, Podcast Addict, and RSS that provides audio narration of the weekly risk brief.²⁰ Episodes are published on a daily cadence and are categorized as News Commentary. No publicly available listener count or download figures have been disclosed. The podcast is narrated by TYPE III AUDIO.¹⁸

Relationship to AI Safety

Sentinel's work intersects with AI safety in two distinct ways.

AI as a monitored risk category: Sentinel treats advanced AI as one of several catastrophic risk domains under active surveillance, assigning probabilities to AI-related scenarios to produce structured, auditable signals for a field that often relies on qualitative expert judgment.

AI as an operational tool: Sentinel uses Large Language ModelsConceptLarge Language ModelsComprehensive assessment of LLM capabilities showing training costs growing 2.4x/year (\$78-191M for frontier models, though DeepSeek achieved near-parity at \$6M), o3 reaching 91.6% on AIME and 87...Quality: 62/100 within its detection and summarization pipeline, making it a live case study in AI-assisted risk assessment. This dual role — AI as both subject and instrument — raises questions about reliability and whether AI-generated signals are well-suited to detecting novel AI-related risks specifically.

Team members are active in effective altruism and rationalist communities where AI safety is a central concern, creating overlap with organizations such as MIRIOrganizationMachine Intelligence Research InstituteComprehensive organizational history documenting MIRI's trajectory from pioneering AI safety research (2000-2020) to policy advocacy after acknowledging research failure, with detailed financial da...Quality: 50/100, Redwood ResearchOrganizationRedwood ResearchA nonprofit AI safety and security research organization founded in 2021, known for pioneering AI Control research, developing causal scrubbing interpretability methods, and conducting landmark ali...Quality: 78/100, and the broader technical AI safety ecosystem. This embedding facilitates cross-pollination around threat modeling and forecasting epistemics.

Key Uncertainties

Several unresolved questions bear on Sentinel's long-term utility and credibility.

Detection pipeline efficacy. AI-assisted monitoring may perform well on risk patterns resembling historical precedents but face structural challenges with genuinely novel threats — precisely the scenarios Sentinel aims to prioritize. Whether the LLM-based triage layer would surface unprecedented signal types is not publicly documented.

Signal calibration. The tradeoff between false positives (alarm fatigue) and false negatives (missed warnings) accumulates into reputational and institutional consequences over time. Sentinel has not publicly specified its operating thresholds or the criteria by which they are adjusted.

Comparative validation. No systematic benchmarking against established intelligence, geopolitical, or security organizations has been published. It is therefore difficult to determine whether Sentinel's assessments provide marginal value relative to existing monitoring efforts.

Forecaster network quality. Expanding the forecaster pool risks diluting the epistemic norms — calibration, scope sensitivity, and independence — that give probabilistic risk estimates their credibility.

Funding sustainability. Organizations focused on speculative long-run risks face persistent pressure to demonstrate near-term legible impact. Sentinel has acknowledged this challenge in its public fundraising materials and describes its emergency response team as "somewhat inactive" pending further resources.³

Evidential standards. The field lacks consensus on what would constitute proof that anticipatory catastrophic-risk forecasting is valuable, making it difficult to evaluate Sentinel's contributions or attract sustained institutional support. Sentinel's own impact estimate — 3.5 (0.4 to 11) basis points of existential risk averted at full funding — is produced using internal assumptions and has not been externally validated.⁵

Track record documentation. Sentinel's founders state in funding materials that the foresight team has "a very good track record for short-term forecasts," but no formal track record publication with scoring data has been released as of this writing.¹⁰

Sources