
Is Merlin AI safe for work? Review its browser activity collection, regional privacy policies, page context and shared-content controls before installation.
Merlin AI cannot be labeled safe for every kind of business data. Its privacy policy describes collecting submitted content and, for its enabled browser extension with the required permissions, browsing activity such as URLs and timestamps. Review the policy that applies to your region, restrict extension access and keep sensitive client material out until your organization approves the processing. A product feature or security claim is not blanket authorization.

For the adjacent decision, read merlin ai vs chatgpt plus and merlin.
Merlin's January 20, 2026 privacy policy covers its websites, browser extensions, mobile apps, note-taking app and API services. It directs users in the EEA, UK and Switzerland to a separate European policy. Do not substitute the privacy description of one mobile recording feature for the broader assistant policy that applies to browser and chat work.
Identify the controller, account type and feature your team will use before approving an installation. A browser extension acting on a client webpage has a different data path from pasting a public paragraph into a standalone conversation. Save the policy version you reviewed and ask the vendor for the contractual terms required by your organization rather than treating a general marketing statement as a signed data-processing agreement.
| Decision | What to check | Practical implication |
|---|---|---|
| Extension activity | URLs, timestamps and server-side processing | Review site permissions before enabling |
| Submitted context | Prompts, uploads and conversation records | Classify information before sharing |
| Removal | Stopping collection versus deleting stored data | Request retention and erasure details separately |
The policy's extension section lists activity information including visited URL, previous URL, referrer, timestamp, user identifier, redirects and browser details. It states that activity data is transmitted to servers and stored in backend systems. That is materially different from assuming all extension processing stays on your laptop simply because the control appears inside your browser.
The same section says webpage content is not collected unless you explicitly use a feature that requires processing it, and describes minimizing activity data to support user-facing functions. Read both parts together. They do not justify saying that the extension reads every page's full content, nor do they justify saying no browsing-related information leaves the device. Match permissions to the functions you deliberately enable.
Merlin's policy describes user content such as text inputs, uploaded documents and images, conversation history and feedback. Its product page promotes page-aware assistance, document context and reusable knowledge. Before submitting material, decide whether it contains customer identifiers, credentials, unpublished commercial information or content your client prohibited from entering an external AI service.
A practical review uses public or synthetic examples first. Check which context is selected, where the result is saved and whether a sharing action creates access for someone else. Browser permissions, model-provider processing and shared-content visibility are different questions. Disabling one browser permission is not proof that an existing uploaded file has been deleted from every system involved in an earlier request.
Where your browser permits it, restrict extension site access to approved sites or enable it only when needed. Use a separate work profile to make that boundary easier to maintain, but do not describe a separate profile as cryptographic isolation or a replacement for company policy. Inspect the extension's current permissions after an update or a change in the features you use.
Merlin's policy says disabling or removing the extension stops further extension activity collection. That does not establish immediate deletion of already collected records, chat history or account information. If you need erasure or a retention commitment, follow the policy's contact and rights process and obtain the answer relevant to the account and region. Keep the distinction between stopping collection and deleting prior data explicit.
The right business question is whether this specific deployment is appropriate for a defined class of work. An internal checklist should name the allowed data, prohibited sites, approved users, review owner and procedure for removing access. If a regulated or client-confidential workflow needs a particular residency, retention or provider restriction, ask for evidence before enabling it.
We did not conduct a penetration test or independently audit Merlin's certifications. We also do not infer model-training exclusions for every provider from a generic security heading. The policy provides useful evidence about described collection, but an organization's legal and security requirements still need a separate review. When those requirements cannot be established, use redacted data or do not process that task in the service.
A browser extension is not the only way to organize multi-model work. Krater's paid workspace combines 500+ AI models and 1,400+ app connections, but it also requires an appropriate data review for the task. Compare the actual privacy controls and selected model settings instead of assuming that moving vendors removes every processing obligation.
No. Its extension policy describes server-side processing and storage of specified browsing activity.
The reviewed policy says removal stops further extension activity collection. That is not the same as a verified deletion of all previously stored records.
Only after checking your obligations, applicable policy and required contractual controls. This article does not grant that approval.
Merlin official policy and product text checked November 14, 2026 through web retrieval after direct requests returned 403. The European policy was identified as the applicable regional alternative, not audited clause by clause here. No current plan limits, blanket security certification or universal no-training guarantee is asserted.