Public AI Governance Documentation

AI Governance Changelog

Public change log for HireProxy's AI governance package.

2026-10-14

• The job description report's list of requirements is now read from the job description alone, in a separate step that never sees the candidate's profile, and the same list is used for every candidate and every run. Before, the model chose the list while also judging the profile, so the number of requirements, and therefore "found in N of M", could change between runs on the same posting and could depend on the candidate.
• Tightened the definitions of the three degrees of evidence ("found", "partial", "not found") and run the report at a higher reasoning effort, so the same profile and posting produce the same counts more often. Measured on four postings, six runs each: the headline moved in 12 of 24 runs before and 7 of 24 after, none of them because the requirement count changed.
• The requirement list follows the job description's own list of qualifications, one entry per item in its order, then the responsibilities that define the role, so a posting's stated requirements are all represented and none is split into sub-points or dropped.
• The wording "verified" is rewritten to "confirmed" in everything shown to a visitor, because HireProxy does not verify candidate-supplied information.

2026-10-13

• Changed the job description report so that it describes and does not rate. It no longer produces an overall label ("Strong Fit", "Moderate Fit", "Not a Fit"), a closing paragraph advocating for the candidate, or red/amber/green indicators. Each requirement now shows the degree of evidence the profile contains, the headline is a count of those degrees, and the model's summary is descriptive and balanced.
• Added to the report instructions: do not consider or mention protected characteristics or anything that stands in for them; use neutral wording with no praise or criticism of the person.
• Added a disclosure to every report (on screen and in the PDF) that it is AI-generated from candidate-provided information, is not a hiring recommendation, and that a person decides.
• Reports saved before the change are shown in the new format; their overall label and closing paragraph are no longer displayed.

2026-07-09

• Updated Interview Mode rehearsal evaluation criteria to emphasize question relevance, answer clarity, candidate-owned action, result or impact, and story fit.
• Kept quick spoken coaching concise and scoreless while preserving the full written rubric as the scoring source of truth.
• Clarified the free voice-practice cutoff and Career Pro upgrade path while preserving typed rehearsal fallback.

2026-07-06

• Created public AI governance package.
• Mapped HireProxy to NIST AI RMF 1.0 Govern, Map, Measure, and Manage functions.
• Mapped HireProxy to the NIST AI 600-1 Generative AI Profile risk categories.
• Added public system cards for career-agent, fit-analysis, interview-prep, voice-practice, extraction, and career-copy assistance surfaces.
• Added a solo-founder evaluation and monitoring cadence.
• Added an AI incident register template.