Playbooks · Coding tasks
Claude Code 2.1.325 widened the email check
Written November 15, 2026
2.1.325 expands the account-email check to scripts a command runs, file writes, web fetches and MCP calls. What it refuses now, in the notes' words.
The line in 2.1.325
The whole change is one line in Anthropic's Claude Code changelog, under 2.1.325: "Improved the account email check: it now also refuses your email as a client's identity in scripts a command runs, file writes, web fetches and MCP calls." Nothing else in the entry expands on it — no setting, no exception list, no new prompt — so the scope of the check is exactly the four surfaces that line names, plus the places it already covered. If you use Claude Code with an account email you care about keeping out of other systems, this line changes what the tool will and will not do on your behalf.
How the check got here
The check exists because the address could leak. The original exposure — Claude Code sending the account email to websites, covered in Claude Code sent your email to websites — led Anthropic to add a guard: the 2.1.317 entry in the same changelog "added a check that asks before a web or shell tool call sends your account email address to the network." An earlier entry, 2.1.234, had already told Claude to use your account email only to identify you, and not to send it to unrelated services unless you ask. The guard then changed shape. In 2.1.321, covered in The Claude Code email check got narrower in 2.1.321, Anthropic "changed the account email check to refuse only a shell command that holds your email and names a User-Agent, without asking, so other commands and scheduled runs no longer stop at its prompt." The check got narrower and quieter: one pattern refused outright, everything else let through without a prompt.
What it refuses now
2.1.325 widens the refusal again, and the wording matters. The check "now also refuses your email as a client's identity" in four places: scripts a command runs, file writes, web fetches, and MCP calls. Read plainly, the target is impersonation rather than transmission: your account email presented as who a client is — an identity a script claims, a value written into a file as the client's identity, an identity sent with a fetch, or one handed to an MCP server. That is broader than the 2.1.321 pattern, which needed both the email and a named User-Agent inside a single shell command before it refused. A script that builds a request, a tool that writes a config file, or an MCP call that identifies its client can each trip the check now, whether or not a User-Agent appears anywhere. Workflows that deliberately use your account email as an identity string — a script that authenticates as you, a generated file that records you as the client — are the ones most likely to meet a refusal after this update.
What the check still is not
The widening is a set of refusals at the points where the email would be used. It is not a removal of the address from the session. The changelog's own framing across these entries is consistent: Claude is told to use the email only to identify you (2.1.234), a check asks or refuses when a tool call would send it onward (2.1.317, 2.1.321, 2.1.325), and nothing in the 2.1.325 entry says the address stops being available to the model in the first place. Treat the check as a guardrail on outputs, not as privacy from the tool itself. Anyone whose concern is that the account email must never enter the session at all gets no new promise from this release; anyone whose worry is that the email should not be shipped to third parties as an identity gets a wider net than last week.
The rest of 2.1.325, in one line
The same release also fixes a zstd response-decoding security issue, gateway sessions failing on context length, and background sessions drifting onto the wrong sign-in; those changes are covered in Claude Code 2.1.325 fixes a zstd security issue and are not repeated here.
What to do
Update to 2.1.325, and expect the refusals if your automation presents your account email as a client identity in scripts, files, fetches, or MCP calls. Where a tool lets you set a separate identity — a service account address, a project alias, a named client — switch to that before the refusal finds you mid-run. If a legitimate workflow stops, the changelog names no override setting, so the practical fix is changing the identity the workflow presents, not arguing with the check.
Related playbooks
- Check a Cursor machine for the tampered Claude Code extension
- Claude Code 2.1.325 fixes a zstd security issue
- Claude Code sent your email to websites
- Check whether your Claude Code hooks are adding tokens to every prompt or cutting them
Sources
- Claude Code CHANGELOG — GitHub, anthropics/claude-code
- Claude Code v2.1.325 release — GitHub, anthropics/claude-code
Back to: Coding playbooks hub · Also: Claude Code sent your email to websites · The Claude Code email check got narrower in 2.1.321