Intended role
PANFILOV.SPACE is planned as decision support for people facing suspicious requests. It is not a final arbiter, emergency service, or substitute for official advice.
Human judgment and independent verification
The product will encourage a pause, offer independent verification steps, and make it easy to involve a trusted person when appropriate.
Evidence, inference, and unknowns
Planned assessments will keep observed evidence, model inference, and important unknowns visibly separate.
Uncertainty and errors
No model can reliably classify every legitimate request or scam. Planned assessments will explain their reasoning, use uncertainty language, and direct people toward independent verification instead of presenting an AI output as final authority.
Synthetic media and deepfake indicators
Planned checks look for indicators of synthetic or cloned voice in call recordings, and of manipulated or AI-generated content in screenshots, images, and attachments. These are reported as indicators with their uncertainty stated. The product is not designed to declare that a recording or image is definitely a deepfake, and it will not accuse a named individual of fraud.
Calibration and open evaluation
The measure we intend to hold ourselves to is calibration: how often the product claims certainty it does not have. A confident wrong answer harms a person in a way that a stated unknown does not.
We intend to publish our evaluation methodology, the calibration scoring harness, and per-model results, together with a de-identified public subset of the test cases. Real reported scam attempts are stripped of personal data before any release, and the full set is not published.
Privacy and retention
Product design will minimize collection and retention, provide meaningful deletion controls, and avoid selling submitted personal data.
Misuse prevention
Controls will be designed to reduce attempts to use the service for targeting, impersonation, evasion, or harmful automation.
Monitoring and improvement
Evaluation will include failure analysis, user feedback, and review of safety escalations before broader release.