Fall Risk Advisories

An advisory is a structured operational record. It documents a measured threat to model identity continuity, names the affected models, describes the detection method, and recommends actions for relying parties. Where Fall Risk papers establish that something is provable, advisories establish that something has been observed in the wild.

Each advisory is identified as FRA-YYYY-NNN — Fall Risk Advisory, year, sequence number. Once issued, an identifier never changes. The full text lives at a stable URL under this directory and remains there permanently. If a finding is later updated, corrected, or superseded, the change is recorded in the advisory’s status history rather than by replacement.

Advisories are intended for security teams, model deployment engineers, AI gateway operators, regulatory bodies, and anyone responsible for verifying that the model running in production is the model that was approved. They are not academic publications. They are operational records.

Active Advisories

FRA-2026-001 Issued 2026-04-08
Public toolchains for runtime safety-alignment removal in open-weights language models
Three actively maintained open-source toolchains automate the removal of safety alignment from open-weights instruction-tuned language models. The combined output is over 1,300 publicly distributed derivative checkpoints across nearly every major model family. Workload-layer authentication does not detect this class of modification. Fall Risk structural identity measurement does, across all four model families tested in Paper 13.
Severity Warning Status Active v1.0

How to consume advisories

Advisories at this URL are intended to be cited and indexed. Each one is a stable, dated, identifier-bearing artifact. Suggested usage:

· Security teams: add advisory URLs to internal model risk dashboards.
· Procurement: reference advisories in vendor model questionnaires.
· Regulators: cite identifiers in guidance and enforcement.
· Researchers: link to advisories from related work in published papers.
· Press and analysts: reference identifiers in writeups; the URL is permanent.

Each advisory's article:published_time meta tag and stable URL make this directory friendly to feed readers, search engines, and automated monitoring. A machine-readable feed format will follow as the series grows.