BlackFog expands ADX Vision with controls for AI agent prompts
BlackFog introduces prompt checks and optional auditing as ADX Vision expands into AI agent workflows
BlackFog announced ADX Vision 2.0 on September 30. The company says the software is available for Windows and macOS, extending its endpoint controls to prompts sent by employees and AI agents. BlackFog puts its controls on the device before data reaches an AI service.
What administrators can control
According to the launch announcement, seven categories of checks target manipulated prompts, including obfuscation and jailbreak attempts. It also describes routing requests to approved AI services, optional auditing of frontier model prompts and token usage reporting across models.
BlackFog says usage reporting could help identify token theft. The announcement provides no results establishing how reliably it detects abuse.
Its product page says administrators choose what information counts as sensitive, which activity is recorded, how long records are retained and who can view them. Access to those records is governed by roles.
The launch builds on earlier releases
ADX Vision itself predates this announcement. The Windows change log lists model discovery and governance in version 5.5.0, dated December 1 2025. Its latest listed build is 5.6.0 from September 17 2026.
The macOS change log lists support for detecting frontier models in version 5.5.0, dated May 27 2026. BlackFog separately announced general availability for macOS on June 18. These dates describe earlier foundations, not the full September package.
The enterprise change log says current release notes are available inside the console. Its public entries are archived. Buyers should ask which installed agent versions and console settings enable each advertised capability before treating the launch name as a deployment checklist.
Test the workflow and the record it leaves
Start an evaluation with harmless sample data. Try an ordinary request, a request containing a made up sensitive value and a manipulated instruction. Check what gets blocked, what reaches the intended service and what appears in the audit record. Repeat the tests in the actual browser and agent workflows the team uses.
The useful result is a documented boundary for a specific setup. Teams also need an owner for exceptions and a clear answer about who can read retained prompts. ByteForward has not tested the software. The vendor materials reviewed here do not establish a general guarantee against prompt injection or data loss.
Archival photograph Desk Setup by Luke Chesser, dated October 30 2014, via Wikimedia Commons. Released under CC0 1.0 Universal. The photograph illustrates a desktop workspace and does not show BlackFog software.
