Ulterior Safety

Technology

A monitor that reads the model, not only its words.

Most safety checks judge what a model writes. Ulterior Safety’s detectors read what the model is doing inside while it writes, so they do not depend on the model’s own safety training.

01Detector family

Ghost

Current edition: Ghost P

Ghost watches a model’s internal activity as it answers and raises a flag when a request is harmful.

  • Monitors the model’s internal activity while it answers, not only the finished text.
  • Flags harmful requests before the reply is finished.
  • Keeps working when the model’s built-in safety has been removed.
  • Ghost P caught 95.4% of clearly harmful requests at 1% false alarms, on held-out public tests. See the results.

02Next generation

Wraith

In development

The next generation of our detectors. In development.

03Research tool

Brain Lab

Internal, not public

Brain Lab is the internal tool Ulterior Safety uses to explore what happens inside a model while it answers. It supports the research behind the detectors and is not offered publicly.

04Who it is for

Teams that cannot rely on the model’s own safety.

Labs and teams running open models who need a monitor that does not depend on the model’s own safety training.

  1. AI labs

    Evaluating and testing open models, including copies whose safety training has been altered.

  2. Teams deploying open models

    Running models on their own infrastructure, where an independent check on harmful requests helps.

Ghost is at research stage. There is no public API or product release yet; a deployable monitor for self-hosted and open models is on the roadmap.

Talk to us about evaluating Ghost on your models

Contact

Talk to Ulterior Safety.

Investors, AI labs, teams deploying AI and press: we would like to hear from you.

Get in touch Email us