Claude is positioned as the “safer, more thoughtful” AI assistant. Anthropic emphasizes its Constitutional AI approach and responsible deployment. What the company is less vocal about is its data retention practices. I went through Anthropic’s privacy policy and terms of service to understand what actually gets logged when you use Claude.claude.ai. The short version: more than most people assume, and the defaults are not privacy-friendly. If you’re using Claude for client work, competitive research, or anything that should stay confidential, you need to understand the data model before you start typing.
The short answer
Anthropic collects your conversations, IP address, browser fingerprint, and account details when you use Claude. Conversations may be used to train future models unless you opt out via a support request — there’s no in-app toggle as of mid-2026. A VPN masks your IP and encrypts your traffic at the network level, preventing your ISP from logging your AI sessions and stopping Anthropic from tying your prompts to your real network location. It doesn’t prevent Anthropic from storing conversation content, but it removes a significant metadata layer.
What Anthropic collects when you use Claude
Anthropic’s privacy policy distinguishes between “usage data” and “conversation content.” Both are collected by default.
Usage data includes: IP address, browser type, device identifiers, timestamps, session length, and feature interactions. This is the metadata layer — it tells Anthropic where you’re connecting from, how often, and for how long.
Conversation content includes: the full text of your prompts and Claude’s responses. Anthropic states this may be reviewed by employees for safety purposes and used to improve models. The review clause is standard across AI companies, but the training data use is the more significant concern for professional users.
Account data includes: your email address, any payment information, and API usage if you’re on a developer plan.
The API is different. Claude’s API terms specify that API inputs and outputs are not used for training by default, which is why enterprise teams should default to API access over the web interface. If you’re a solo user on claude.ai, that protection doesn’t apply.
VPN protection layer for Claude users
NordVPN addresses the network-level exposure. Here’s what changes when you route Claude traffic through NordVPN:
IP masking. Your real IP is replaced with a NordVPN server IP. Anthropic’s logs show the VPN server location, not your actual location. With 6,400+ servers across 111 countries, you can choose jurisdictions with strong privacy laws.
ISP visibility. Without a VPN, your internet provider can see that you’re making HTTPS requests to anthropic.com. With NordVPN active, they see only encrypted traffic to a VPN endpoint. Your AI session usage is not visible in your ISP’s logs.
Traffic encryption. NordLynx (WireGuard-based) provides modern encryption with minimal overhead. This matters if you’re on shared networks — coffee shops, coworking spaces, hotel WiFi — where traffic inspection is a real risk.
DNS-level protection. Threat Protection Pro blocks trackers and malicious domains before they reach your browser. When you visit claude.ai, third-party analytics scripts often load alongside the main interface. Threat Protection Pro stops those from building a behavioral profile.
The practical setup for Claude:
- Enable NordVPN and select a privacy-favorable jurisdiction (Netherlands, Switzerland, Iceland)
- Use NordLynx protocol for best performance
- Enable Threat Protection Pro in the NordVPN app
- Open Claude in a fresh browser tab — the session now originates from the VPN IP
NordVPN’s RAM-only server infrastructure means there are no persistent logs of your VPN activity. Even if Anthropic received a legal request for your IP’s history, the VPN server has nothing to hand over. This is backed by independent audits from PwC and Deloitte.
Managing Claude’s data retention directly
VPN handles the network layer. You also need to manage data retention within Anthropic’s platform:
Opt out of training. Anthropic allows users to request that their conversations not be used for training, but as of mid-2026, this requires a manual support request rather than an in-app toggle. Submit a request through their privacy portal.
Delete conversation history. You can delete individual conversations or all history from Settings. Deletion removes conversations from your view; Anthropic’s retention policy specifies they may keep data for up to 30 days after deletion for safety audits.
Use temporary projects. Claude’s Projects feature lets you create isolated workspaces. While this doesn’t change data collection, it helps compartmentalize what you share.
Enterprise and Teams plans. Anthropic’s enterprise agreements include enhanced data protection provisions, including stricter limits on employee access to conversations. If you’re using Claude for business, the enterprise plan is worth evaluating.
Get NordVPN
NordVPN is the VPN I run on all my devices when using AI tools. At $3.39/month on the 2-year Basic plan, it’s the lowest-friction way to add network-level privacy to your AI workflow. It supports 10 simultaneous devices, so one subscription covers your laptop, phone, and any other machines you use for AI work. The 30-day money-back guarantee makes it risk-free to test.
Recommended
NordVPN
Encrypt your AI chats, mask your IP across geo-restricted models, and keep client data private across 60+ countries.
FAQ
Does Anthropic share Claude data with third parties?
Per Anthropic’s privacy policy, they may share data with service providers (hosting, analytics), law enforcement when legally required, and in the event of a business acquisition. They do not sell user data to advertisers.
Is Claude safer to use than ChatGPT from a privacy standpoint?
Both platforms collect similar categories of data. Claude’s API has slightly more explicit training-data exclusions by default. Neither is “private” in the way a local model would be. For network privacy, the same VPN setup applies to both.
Can Anthropic employees read my Claude conversations?
Yes, Anthropic’s policy states that conversations may be reviewed by trained employees for safety and quality purposes. This is standard across AI labs. If you’re handling genuinely confidential information, use the API with appropriate data agreements or a local model.
Does Claude work through a VPN?
Yes. Claude.ai functions normally through VPN servers. If you encounter a Cloudflare challenge, switch to a different NordVPN server in the same region.
How do I request that Anthropic delete my data?
Submit a data deletion request through Anthropic’s privacy portal (privacy.anthropic.com). They’re required to respond within 30 days under GDPR and CCPA frameworks if you qualify under those jurisdictions.