Claude
Anthropic's chat assistant. Strong on long documents, and as capable of inventing a fact as any other chatbot.
Never for sensitive material
Never put this into it
Do not type, paste, upload, or dictate:
- The real name of any witness, survivor, informant, or activist
- Full witness statements or interview transcripts
- Case files, legal filings, or investigation notes
- Exact addresses, GPS coordinates, or descriptions detailed enough to show where a specific person lives, hides, or works
- Phone numbers, national ID numbers, or passport numbers
- Photos and videos that show an identifiable face if that person's safety depends on staying hidden
Since the model-training toggle defaults to on, treat anything typed here as something a stranger could eventually read, unless you have gone into Settings → Privacy and switched training off yourself.
A safer tool to use instead
Claude is a chat assistant from Anthropic. It does the same work as ChatGPT: drafting, summarizing, translating, answering research questions. Many people find its writing clearer and its handling of long documents stronger.
It invents facts as confidently as any other chatbot. A citation, a date, a legal provision: any of them can come back fluent and wrong. Every output is a draft for a person to verify.
For anything involving a witness, a case file, or a location that has to stay hidden, use a model running on your own computer, such as OllamaOllama (running local models on your own computer)Safe for sensitive materialOpen sourceMITHas a free version: Completely free, no usage limits, no subscription. You need enough disk space and RAM for whichever model you choose to run, several gigabytes for a usable model.View tool.
What happens to your data
Where this data goes
- Anthropic servers (United States)
Trains on what you type?
By default yes, but there is a setting to opt out.
How long it is kept
Settings → Privacy holds the training toggle, and it arrives pre-set to allow training: anyone who does not go in and switch it off is training the model by default.
Since 28 August 2025 every Free, Pro, and Max account is asked whether chats and coding sessions can be used to train future Claude models. Allowing training extends retention to five years; declining keeps the older 30-day retention. Safety-flagged conversations may be kept regardless of this setting. Business and API usage sit under separate commercial terms that exclude customer content from training.
Account required
- An email address
- A phone number
Getting it, and using it in Iraq
Access in Iraq
- Paid plans typically need a card issued outside Iraq.
- Signing up needs a phone number.
Iraq is on Anthropic's own published list of supported countries and territories for both Claude.ai and the API: the service itself is not geo-blocked. Payment is the practical obstacle organisations in the region report: Iraqi-issued cards are widely reported to be declined by international SaaS billing, a pattern Anthropic does not document, so budget for a card issued abroad if you need the paid plan. Every account, free or paid, requires a working phone number for SMS verification at signup. There is no way to sign up with only an email address.
Language quality
No one has published a measurement for this.
Arabic (Modern Standard)
Anthropic's published model and system cards were checked for this entry and contain no per-language accuracy breakdown for Arabic, and no independent benchmark was run. Untested.
Iraqi Arabic
No dialect-specific benchmark for Iraqi colloquial Arabic exists or was checked for this entry. Untested.
Kurdish (Sorani)
No benchmark or vendor documentation on Kurdish Sorani performance was found or checked for this entry. Untested.
Sources
Last checked: 2026-08-31