The settings in these guides disable optional data collection and experimental features.
Here's exactly what that means, in plain language.
In Claude Code (CLI and VSCode)
Your settings.json includes these lines:
| Setting |
What it does |
CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC |
A master switch that turns off usage telemetry, error reporting, the feedback command, and automatic updates — all in one. |
CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY |
Suppresses the occasional "How is Claude doing this session?" survey prompt. (Not covered by the master switch above, so it's listed separately.) |
CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS |
Turns off experimental/beta features. As a bonus, it also improves compatibility with gateways like Cornell's that can reject experimental request headers. |
In Claude Desktop
IT applies these settings for you as part of the managed configuration — there
is nothing for you to turn on:
| Setting |
What it does |
| Block nonessential telemetry |
No product-usage analytics leave your computer. |
| Block nonessential services |
Stops optional cosmetic web fetches. Note: artifact previews and connector icons won't render in conversations. |
| Block essential telemetry |
No crash/error reports leave your computer either. The most private setting; note that it also removes remote crash visibility if you ever need support. |
What still connects to the internet
Even with everything above turned off, two destinations remain necessary:
api.ai.it.cornell.edu — Cornell's gateway, where your prompts are actually
answered.downloads.claude.ai (Claude Desktop only) — used once at startup to load the
Cowork workspace.
Your prompts go to Cornell's gateway, governed by Cornell's policies — not to a
personal Anthropic account.
The Code tab
Claude Desktop's Code tab is off by default and enabled per person on
request. Some people in the pilot have it and some don't — it depends on what
their work actually calls for and how comfortable they are with the way it
operates. There is no single answer for everyone, and not having it isn't a
judgement about you. If it's off, you'll see a dismissible "Code has been
disabled by your organization's administrator" message on first launch.
Why it isn't simply on for everyone
Chat and Cowork answer you. Code acts: it reads and writes files on your
machine, runs commands, and carries multi-step work forward with much less
checking in along the way. That is exactly what makes it useful, and it is a
genuinely different proposition from a chat window. The things worth
understanding first are just the ordinary consequences of that autonomy:
- It can create, change, and delete files — and can work through them faster
than you can read what it's doing.
- It runs commands with your access. It asks before most actions, but the
value of it comes from handing over more of the loop than you would in chat,
so the prompts get easier to wave through the longer you work.
- It can act on content it was given, not only instructions you typed — a web
page or document it reads may contain text written to steer it.
- Long autonomous sessions are the most expensive way to use the gateway, and
they move your actual spend faster than chat does — see Getting Your Cornell AI Gateway Key.
None of that makes it dangerous so much as not self-explanatory. The default
exists so people meet Code deliberately — knowing what it can do for them and
what it can do without asking first — rather than discovering both at once on
a live piece of work.
Asking for it
Email itrequests@business.cornell.edu. It's a per-person setting rather than
an approval process, so there's no case to argue and nothing to justify — we'll
switch it on, and talk through any of the above you'd like to talk through.
If you'd rather try the same capability without changing your Desktop setup,
Claude Code on the command line is the same idea in a different wrapper —
see Appendix: Claude Code (CLI) on Cornell's AI Gateway.
Questions about data handling or the gateway? Email itrequests@business.cornell.edu.