SUPERSTRATEGY – probabilistically optimised, frontier-forecasted decisions.

Cassi is a Superstrategy engine that forecasts 50% more accurately than humans, combining AI and collective intelligence to improve decisions.

Media coverage

Who we are. Why we exist.

Cassi pairs collective intelligence, your team's own judgement, made legible and stripped of bias, with frontier AI forecasting. Together, they're more accurate than either alone.

In Greek myth, Cassandra could see the future perfectly. Nobody believed her.

Cassi is built to fix that curse for organisations. It tells you what you need to know, not what you want to hear.

Benchmarks · Live

#1AI on the ForecastBench leaderboard
  1. 1ForecastBenchSuperforecaster median forecastElite humans69.2
  2. 2CassiCassi-2026-05-10#1 AI68.8
  3. 3xAIGrok 4.20 (Beta, C)68.1
  4. 3xAIGrok 4.20 (Beta, D)68.1
  5. 5Google DeepMindgreen tree67.5

23 tournaments entered

ForecastBench leaderboard last updated: 17/07/2026 at 07:31 am

The Forecast Room

Cassi puts numbers on the AI scenario reports everyone's talking about — every resolvable claim, priced and tracked.

11%

AI causes a major decline in UK financial and professional services

Likelihood over time

Aug 2026Sept 2026
11%

Question: Will employment or gross value added in the UK's financial and professional-services sectors fall at least 10% from 2026 levels, with AI identified as a major cause, by the end of 2030?

Resolves: Will resolve as true if, by 01 Jan 32, ONS data show at least a 10% decline from the 2026 baseline in employment or real gross value added across the specified sectors by end-2030, and an official or well-supported independent analysis identifies AI as a major cause. It will otherwise resolve as false.

Resolution date: 1 Jan 2032

53%

AI reliably completes month-long cognitive tasks

Likelihood over time

Jul 2026Sept 2026
53%

Question: Will frontier AI systems reliably complete cognitive tasks that would take an experienced human a month or longer, before 1 Jan 31?

Resolves: TRUE if METR-style data or an authoritative assessment (AISI / International AI Safety Report) documents reliable completion of ~month-long tasks; otherwise FALSE.

Resolution date: 1 Jan 2031

67%

Frontier capabilities outpace safety evaluation

Likelihood over time

Aug 2026Sept 2026
67%

Question: Will a leading AI-safety authority state that frontier capabilities cannot be adequately evaluated before deployment because evaluation methods have fallen behind, before 1 Jan 31?

Resolves: Will resolve as true if, by 01 Jan 31, the UK AI Security Institute, the International AI Safety Report or a comparable independent authority explicitly states that existing pre-deployment evaluation methods are materially inadequate because frontier capabilities have outpaced them. It will otherwise resolve as false.

Resolution date: 1 Jan 2031

40%

Large anti-AI protests or AI as a decisive election issue

Likelihood over time

Jul 2026Sept 2026
40%

Question: Will large-scale anti-AI public protests occur, or AI become a decisive issue in a major-democracy national election, before 1 Jan 31?

Resolves: TRUE if either branch is satisfied before 1 Jan 2031. Protest branch. A single demonstration against AI development or AI infrastructure draws 25,000 or more participants per an independent estimate — police, a national news organisation's own count, or the Crowd Counting Consortium — or coordinated demonstrations occur in ten or more cities across three or more countries within a 30-day period with a credible aggregate of 50,000 or more. Does not count: organiser-supplied crowd estimates; superlative characterisations ("largest ever," "first major") unaccompanied by a participant count; single-outlet or student-newspaper sourcing; online petitions, open letters, or strike ballots. Election branch. In a national election in a country with population above 30 million rated a full or flawed democracy by the EIU Democracy Index, AI — including AI-driven job displacement, energy costs, or data-centre siting — is named among the top three issues by voters in at least two independent national polls conducted during the formal campaign period, and receives substantive treatment in the platforms of parties taking a combined majority of votes cast. Does not count: AI appearing in manifestos without polling salience; commentary asserting AI was decisive absent polling evidence; sub-national or single-constituency contests. Otherwise FALSE.

Resolution date: 1 Jan 2031

Enter The Forecast Room