About Akaeon
Building the infrastructure for an AI-forward world.
AI is being deployed faster than the substrate underneath it can support. Akaeon builds the missing layer — cryptographic, verifiable, neutral — starting with the Akaeon Registry, an opt-out registry for AI training that publishers, labs, courts, and auditors can verify without trusting us.
Why this exists
The dominant interactions of the next decade — what labs train on, what publishers permit, what a model is allowed to say it knows — will be mediated by infrastructure that does not yet exist. Most of it is being improvised inside individual companies. Most of it cannot be audited. Most of it routes trust through whoever happens to be running the server.
That is not a stable foundation for the systems being built on top of it. The fix is not another platform. The fix is shared primitives — cryptographically verifiable, anchored to public substrates, operable without a relationship with the operator. Infrastructure that holds up whether or not we do.
How we operate
Build
We build the primitives the next decade of AI will be transacted on — cryptographic, verifiable, neutral. The Akaeon Registry is the first.
Anchor
Our systems anchor to public, third-party-operated substrates instead of to our own servers. If we disappear, the records still verify. Trust does not route through us.
Steward
Infrastructure is judged on what it does over decades, not quarters. We stay engaged with the systems we ship as long-term operators rather than transient participants.
The current chapter
The Akaeon Registry is a cryptographic opt-out registry for AI training. Publishers register their preferences; labs query a single API at training-data ingestion to honor them with audit-defensible evidence. Every record is independently verifiable against the public Arweave network — no Akaeon code in the verify path, no required ongoing relationship with us.
It runs on the same substrate that powers Stelais’s authorship infrastructure. The registry is the first surface; the substrate underneath it is the longer project.
Research
Building credible infrastructure for an AI-forward world requires understanding how production AI systems actually behave — not how the marketing material says they do. Our in-house research program probes commercial models directly: their fingerprints, failure modes, and the limits of what current detection and provenance schemes can guarantee.
Recent work includes the first empirical documentation of reproducible multi-modal attractor sampling in a commercial multimodal image-output model, and categorical behavioral separation of three production image-to-image APIs. The findings feed directly back into how we design the substrate. See /thinking for the full papers.