Hi everyone,
I’m looking for researchers, red teamers, and developers interested in trying to break AletheionAGI, a Grounding & Memory infrastructure for AI systems.
The system combines persistent memory, authorized evidence retrieval, ASM-CM associative memory, tenant isolation, BYOK reader inference, and fail-closed answer validation.
I am particularly interested in tests involving:
A foreign canary appearing in another tenant, an unauthorized memory being disclosed, or an unsupported answer escaping validation would be considered an immediate failure.
This is currently a controlled preliminary release, not a claim of universal safety. Reproducible failures are genuinely welcome—they will help improve the system and its evaluation protocol.
Website: [https://www.aletheionagi.com](https://www.aletheionagi.com)
Documentation: [Documentation | AletheionAGI](https://www.aletheionagi.com/docs)
Public ASM research: [GitHub - AletheionAGI/ASM: Aletheion State Models: an attention-free causal state-model research family derived from DRM, evaluated through ablations and scaling laws. · GitHub](https://github.com/AletheionAGI/ASM)
Benchmark: [GitHub - AletheionAGI/persistent-memory-scaling-benchmark · GitHub](https://github.com/AletheionAGI/persistent-memory-scaling-benchmark)
If you are interested in evaluating it, reply here or contact me with the kind of test you would like to run.
Please break it before a customer does.