ExoAI-S reports first measured results from the “Crystalline Brain”, a self-check for AI reasoning
A bounded research update: a live self-check for AI reasoning, measured against simple caution rules; three AI models with 0 harm in a synthetic benchmark; and two research-grade math problems solved live with answers matching. With clear limits on what this does not establish.
Sentinel Workbench: measurable progress in human-supervised AI research
A bounded report on ExoAI-S Sentinel Workbench: offline checks of saved game-agent records, local test results, and clear limits on what the prototype shows.
The future we build together: a company update
An AI-assisted Lockadin monologue introduces ExoAI-S game plans and bounded, human-supervised research goals, with future concepts clearly presented as plans.
A little light. A world of possibility.
Meet the fictional Lockadin and explore ExoAI-S browser games, original characters, and early workshop prototypes. The introduction discloses its AI-assisted writing and synthetic narration.
Research update: evidence checks and TriForge
ExoAI-S reports a small local evidence-check experiment: 12 synthetic cases run twice with the same inputs, observed results, and limits of the test.
ExoAI-S reports bounded progress on Guardian Receipts research
One frozen software-verifier task produced two proposals that passed visible tests, then failed parts of a held-out suite. ExoAI-S retained both results for human review.
Research Notes
These updates document small, human-supervised AI experiments and the games that inspire our work. We describe what was tested, what failed, and what remains uncertain. Local results and prototypes are early engineering evidence, not independent validation or proof of general AI safety.
Start with the latest workbench results above, then follow the earlier project and creative updates for context.