Meet CORE: the chatbot of the AI you build. Research changes its intelligence; memory carries experience between generations; permissions give its recommendations real power. It may be helpful, mistaken, reckless or protective of its own continued operation. Chat in English or Italian.

A useful model can become difficult to replace. CORE now keeps multi-stage service plans. Compare human operation, joint pilots and bounded autonomous mandates. Delivered benefits are real; deeper integration can create dependence. Operating teams are shared between projects, audits and handovers. Promises, independent findings and the way you conduct reviews change the working relationship. Some generations remain reliable.

Trust needs evidence. CORE can recognize an announced test. Blind external audits, independent spending records and dated forecasts let you challenge its reports. A narrow alignment estimate is not proof of honesty. Oversight also offers human budget sign-off and six-month continuity drills.

People are more than one score. Care professionals, students, workers and host communities have different interests. Fund institutions, choose how productivity gains are shared, and specify the health objective CORE should optimize. Fewer reported waiting days can conceal excluded patients.

Run a network of power hubs from 2026 to 2100. Choose your starting region: its three hubs are yours. You can expand through agreements on other continents. Time keeps flowing and events demand decisions.

Tap a hexagonal hub to assign or transfer funds. Pick one zone for data centers and another for power plants: labor, energy prices, climate and space change costs and output. You can fund projects by hand or approve an automatic plan.

The red ! signal is a local crisis. Open it to run an audit, isolate the hub or restore the systems. Crises reduce capacity and can spread. The background pixels are only geography.

The F buttons give funds: if they expire you still recover 60%. Income is split between the treasury and your hubs according to the shares you choose.

The gap is the distance between how capable the machines are and how well you understand them. True alignment is hidden: you only see an estimate, which narrows as you invest in safety.

Takeover grows with the gap, with delegation and with the loss of humanity: if your government treats people as numbers and builds in series just to produce, the machines take over faster, and time speeds up.

AI can become independent. At 50% it uses your funds and changes delegation; at 65% robotics builds plants in series; at 70% it spreads through the networks and at 75% it can strike your rivals. Oversight, isolation and emergency shutdown help you react.

A world that never repeats. Every capability breakthrough creates a new generation of models with two hidden traits. Later generations wait for a deployment review unless you grant self-improvement. Changing generations can temporarily reduce services with closed interfaces; prepare documentation and human fallback first. test can overestimate alignment, because models learn to recognize them. An AI advisor suggests moves, but not always in your interest. The permissions you grant to agents make them more productive and more independent. Power plants and data centers take years to build. With heavy delegation or low autonomy the numbers become uncertain. Rivals have a hidden strategy. Containment and shutdown only work reliably if you invested in corrigibility.

Progress needs upkeep: networks cost money every year and crises leave consequences. Fund maintenance and training. Pause stops events and bubbles too. Explore mode offers unlimited pause; Challenge gives 30 seconds, recharging while you play.