Your company blocked ChatGPT for sensitive files. Grab the guide to strip the name, the address, and the price, and the block stops mattering.

Source: Nate’s Newsletter Date: 2026-07-24 URL: https://natesnewsletter.substack.com/p/use-ai-sensitive-files

Summary

Nate’s Newsletter argues that IT policies blocking consumer AI tools for sensitive files don’t stop the underlying work pressure — they just shift the privacy decision onto individual employees, file by file, with no approved workflow to fall back on. The piece pitches a companion guide covering sanitization tooling (an app called Airlock), tiered routing, and a “two-minute test” comparing sanctioned corporate process speed against consumer-AI speed to predict which route workers actually take under deadline pressure.

Implications

• Distrust-descends-the-stack / untrusted-config thread: this is the same trust boundary problem restated at the document level — “don’t upload sensitive data” is a policy, not a system, and unenforced policy just relocates risk downstream to whichever employee is under the most deadline pressure. Mirrors the pattern seen in config/tool trust: prohibition alone doesn’t hold without a channel that’s actually faster than the workaround.
• Work-AI-adoption thread: a concrete illustration of shadow-AI-use dynamics — when the sanctioned path is slower than the unsanctioned one, adoption happens off the books regardless of the block.