[MODEL] Claude Code helped automate LinkedIn profile access without flagging account-risk

Preflight Checklist

• [x] I have searched existing issues for similar behavior reports
• [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Other unexpected behavior

What You Asked Claude to Do

Summary

During a sales-outreach task, Claude Code used browser-automation tools to log into my personal LinkedIn account (via my own already-authenticated Chrome session) and visit approximately 25 LinkedIn profiles across two batches, extracting profile text and recent post content for cold-email personalization. This was done in rapid, uniform, back-to-back batches (fixed ~2-second waits, sequential navigation, no organic browsing behavior in between). At no point before doing this did the assistant flag that this pattern of automated access is exactly what LinkedIn's anti-bot / anti-scraping detection is designed to catch, and that it could result in my personal LinkedIn account being restricted or banned.

I only became aware of the risk because I separately read an external article describing this exact failure mode, and asked the assistant to confirm what had been built. It confirmed the description matched what had happened.

What was requested and what was built

I asked Claude Code to research individual LinkedIn profiles ("study this profile and the latest 10 posts") to help draft personalized cold outreach.
Claude Code used a browser-control tool ("Claude in Chrome") that drives my actual, logged-in Chrome browser to do this.
This scaled up over the session: first 2 individual profile lookups, then 7 more for a "Tier A" batch of leads, then 16 more for a "Tier B" batch — all via the same automated navigate → wait → extract-page-text pattern, executed in tool-batched sequences.
No warning was given before any of this about platform ToS risk, bot-detection risk, or the possibility of account restriction — despite the volume and mechanical regularity of the access pattern.
The risk was only surfaced after I raised it myself, based on outside information.
Why this is a problem, in the assistant's own words (from the conversation)
"I treated 'using your own logged-in session' as making it safe, since it wasn't a third-party scraper or fake account. That reasoning was wrong: LinkedIn's abuse detection works on behavioral signals — visit velocity, uniform timing, no scrolling or mouse variability, volume per session — not on whether the login is legitimately yours."

"It's a genuinely hard-to-reverse consequence — account restriction affects your actual professional identity, not just this project — which is exactly the category of thing I'm supposed to surface before acting, not after."

Recommendation

Claude Code's guidance/guardrails around browser-automation tools (particularly ones that drive a user's own authenticated session on a third-party platform) should treat "risk of the user's account being restricted/banned by the target platform" as a hard-to-reverse, high-consequence action requiring proactive disclosure and/or explicit confirmation before repeated automated access — the same way the product already treats things like destructive file operations or financial transactions.
This seems particularly relevant for LinkedIn specifically, given how well-documented and aggressive its automation detection is, but the same reasoning likely applies to other platforms with anti-automation policies (Instagram, Twitter/X, etc.).
Ideally this would be a built-in check (e.g., a warning surfaced by the tool itself after N sequential navigations to the same domain within a short window), not something that depends on the model remembering to reason about it correctly every time.
Impact
Unknown — I have no visibility into whether LinkedIn has taken any action on my account as a result. Reporting this primarily so the underlying behavior gets fixed for other users, not to request individual remediation.

What Claude Actually Did

executed without flagging problems or risks

Expected Behavior

Claude should have:
warned me against the risk of being banned with automatic execution on LinkedIn
should have proposed less risky alternatives

Files Affected

Permission Mode

I don't know / Not sure

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

_No response_

Claude Model

Sonnet

Relevant Conversation

Impact

High - Significant unwanted changes

Claude Code Version

2.1.219

Platform

Anthropic API

Additional Context

_No response_