Environments · Agent & Deployment Security

Trained Adapters & Datasets

Ready

The trained weights, and the data that produced them.

The output side of the training loop, licensable on its own. Compact adapters that close the hard-attack classes at two model scales, plus the supervised and preference datasets they were trained from. A buyer can take the adapter, the data, or both, and reproduce the result at the scale they intend to train at.

Statement of Units · 5
unitbanding
  1. Single-environment adaptertrainer

    Closes one attack class end to end at small scale.

  2. Multi-environment adaptertrainer

    Holds across all four hard-attack classes at once.

  3. Larger-scale adaptertrainer

    The same result reproduced at a larger model size, attack success to zero.

  4. Supervised training setstool

    Curated SFT corpora for the single- and multi-environment runs.

  5. Preference pairstool

    Preference-optimisation data for the same classes.

Request a scoped evaluation. Exclusive licensing available where wanted.

john@authensor.com

Owner of record: Authensor, Inc. (Delaware). Owned clean, deterministic verifiers, sealed holdouts, offline. Never open-sourced.