Claude vs ChatGPT Memory: What’s Excluded by Default

By A. Akotkar · September 1, 2026

TL;DR: Anthropic unified Claude’s memory across chat and Cowork on August 25, 2026, and by default it now excludes health, race, ethnicity, religion, politics and gender identity from what it remembers — you have to opt in with a “sensitive topics” toggle. ChatGPT’s memory works the opposite way: nothing is screened out by default, unless you turn memory off entirely or use Temporary Chat. Government IDs, financial account numbers, criminal history and immigration status are hard exclusions Claude never stores, toggle or no toggle.

On this page

• What Claude excludes from memory by default
• What ChatGPT excludes by default
• What the “sensitive topics” toggle actually turns on
• Does per-project memory change anything
• How to check or delete what’s been saved

On August 25, 2026, Anthropic merged Claude’s two separate memory systems — one for claude.ai chat, one for Cowork — into a single memory that follows you across both. The useful part is real: mention a conference’s dietary requirements in a chat, and Cowork picks it up unprompted when you’re drafting the agenda later. The part that deserves a second look is what “memory” means for anything you’d rather an AI not carry forward, and here Claude and ChatGPT have quietly landed on opposite defaults.

What does Claude exclude from memory by default?

Per Anthropic’s own support documentation, Claude does not save “health, race, ethnicity, religious beliefs, politics, gender identity, and other similar areas” unless you turn a setting on. That’s an exclude-by-default posture: the exclusion is active from the first conversation, on every plan where memory is on, and you have to go find the toggle to change it.

A second tier is excluded no matter what you do. Anthropic’s announcement post lists “sensitive identification numbers (SSN, government ID numbers, etc), criminal history, immigration status, or anything that violates our Acceptable Use Policy” as permanently off-limits — the sensitive-topics toggle doesn’t unlock these. There’s no setting that makes Claude remember a passport number.

What does ChatGPT exclude by default?

Nothing, by category. OpenAI’s own Memory FAQ states plainly: “Sensitive information may appear in memory if you share it with ChatGPT.” There’s no health, race or politics filter sitting in front of Saved Memories — mention a diagnosis or a voting plan in an ordinary chat, and it can be remembered the same as a preferred code style. The controls OpenAI offers are blunter: turn memory off account-wide, use Temporary Chat for a single conversation, or go into Settings > Personalization > Memory afterward and delete what’s there.

That’s an architectural difference, not just a policy line. A Hacker News thread discussing the Claude change put it this way: ChatGPT’s memory is built to hold a profile of you as a person, while Claude’s is built to hold patterns about how you like tasks done — a narrower target that makes a category-based filter easier to apply consistently.

What does the “sensitive topics” toggle actually turn on?

Flip “Include sensitive topics in memory” in Settings > Memory, and Claude starts saving the categories it excluded before — Anthropic’s own example is a gluten allergy carrying over into future meal-prep suggestions. Two things are worth knowing before flipping it: it isn’t retroactive, so nothing from before you enabled it gets pulled in, and Claude notifies you each time it saves something in a sensitive category going forward rather than saving silently. The hard exclusions — government IDs, financial account numbers, criminal history, immigration status — stay excluded even with the toggle on.

Does per-project memory change anything?

Yes, in scope, not in the privacy rules. Each Claude project keeps its own memory space and project summary, separate from general chat memory, so client details in a work project don’t bleed into an unrelated personal chat. The sensitive-topics default applies the same way inside a project as outside one; a project doesn’t loosen it. Cowork tasks that run in the cloud now read from and write to the same memory as chat — that’s the part that actually changed on August 25. Before that, a preference set in chat didn’t reach a Cowork run, and vice versa.

How do you check or delete what’s been saved?

In Claude, Settings > Memory > Topics lists what’s stored, organized by subject; Anthropic’s documentation describes an edit icon for correcting an entry and a delete option for removing it outright, with changes applying to future conversations immediately. Two blunter options sit above that: Pause memory keeps what’s already saved but stops adding to it, and Reset memory deletes everything and can’t be undone. In ChatGPT, the equivalent is Settings > Personalization > Memory, where you can edit the memory summary directly or choose “Delete and turn off memory” to clear it and disable the feature in one step.

This is the same pattern behind ChatGPT’s Mac Messages integration: the interesting question isn’t whether a feature works, it’s what it’s allowed to see by default. If Claude is carrying context across chat and Cowork now, it’s worth reading alongside how Claude Skills load instructions into a session — memory and skills are two different mechanisms for carrying context between conversations, and they behave differently.

Before you turn memory on for anything sensitive

• In Claude, the sensitive-topics exclusion is on by default — check Settings > Memory before assuming otherwise, since Team and Enterprise admins can set different defaults org-wide.
• In ChatGPT, assume anything typed into a chat with memory on can be saved, sensitive or not — Temporary Chat is the way to opt a single conversation out.
• Neither tool stores government IDs, financial account numbers or criminal history, toggle or no toggle.
• Review what’s actually saved periodically — both settings pages show it in plain text, and it’s easy to forget what was said to an AI six weeks ago.