Is Opera Neon safe? What its AI sends, keeps for 365 days, and leaves on your computer

The Morse teamPublished October 30, 202611 min read

As of October 30, 2026, Opera Neon keeps every question you ask its AI, and the pages its agent reads, encrypted on Opera's servers for 365 days, and passes them to OpenAI or Google for answers. That makes Neon safe enough for errands you wouldn't mind Opera holding for a year, and a poor fit for anything behind a login. Opera says your passwords, cookies, and card numbers stay on your computer and that it trains no AI on your data, but Neon Do works inside your signed-in session and can fill in forms and accept consent dialogs for you.

We read Neon's privacy statement (last updated April 9, 2026), its terms of service (last updated December 15, 2025), its FAQ, its subscription page, Opera's security posts and product posts about Neon, and Opera Limited's annual report on Form 20-F for 2025, all on October 30, 2026. Opera earns credit in three places. It gives a separate retention period for each AI provider, it admitted that its own system first rejected a researcher's report of a Neon flaw as a placeholder, and it put the command-line tool that lets your own agents drive Neon on GitHub as open source.

Opera Neon key facts, October 2026

Is this the same Opera Neon from 2017?

No. Opera used the name before, for a "concept browser" it introduced in January 2017 to show off ideas like a visual tab bar and video pop-outs. Today's Opera Neon is an AI agent browser that Opera began shipping on September 30, 2025.

Some results for "is opera neon safe" are about the old one. A volunteer on Opera's own forum calls Neon "an unsupported, one-time experiment/demo" that's "not safe to use from a security standpoint" (Opera forums), and that warning is about the 2017 concept browser. Download the current Neon only from operaneon.com.

Is Opera Neon free?

The browser is, and its AI agents aren't. On August 12, 2026, Opera made Neon a free download so people could connect their own agents to it. The subscription page lists the Free plan with basic AI chat that can summarize and compare pages. The Standard plan costs $19.90 a month, month to month, and adds Neon's agents, deep research, image and video generation, scheduled tasks, long memory, and models from OpenAI, Anthropic, and Google.

Opera's other pages haven't caught up. Its FAQ, read on October 30, 2026, still says "Opera Neon is a paid product and requires a subscription," and its terms still call Neon "a paid service." Neon's privacy statement doesn't treat the two plans differently, and even the Free plan's chat sends pages to the cloud.

Does Opera Neon run locally or in the cloud?

Mostly in the cloud. Neon Do runs in your browser, and Opera's FAQ says it still "uses cloud-based large language models (LLMs) to generate the plans and instructions it follows." Here is each part of Neon, from Opera's own pages, read October 30, 2026.

Neon Do reads web pages through their code rather than screenshots "as much as possible," Opera's FAQ says, using "the structure of the webpages themselves (DOM tree and layout data)." Opera says the credentials you type into those pages are kept out of it.

Which AI companies see what you ask Neon?

OpenAI and Google, by name, plus the provider of any other model you pick in Neon Chat. Neon's privacy statement says your input "is shared with OpenAI or Google (both US-based companies)" to answer it. The same statement says OpenAI deletes it after 30 days and Google after 24 hours, and that neither uses it to train models. Opera's FAQ adds that Opera's agreements with providers such as OpenAI and Google "prohibit them from using Opera users' data to train their models."

The model menu reaches further than that list. The FAQ says Neon Chat offers "Gemini, Grok, GPT and GPT Pro, Claude Opus and Claude Sonnet, Deepseek, GLM-5 and Qwen3." The privacy statement's list of technology providers names only OpenAI and Google among AI companies, and Opera's January 29, 2026 post says even Meta's Llama 4 Maverick in Neon is "hosted by Google." For other Neon Chat models, it says, "providers of such AI models and their respective privacy policies will be displayed," so check the model picker before you choose Grok, Claude, or DeepSeek.

Google also gets some prompts for a second reason. The privacy statement says Neon picks out "categories of prompts" to send to Google for search suggestions, "masked so as not to be associated with your Google account or your Neon/Opera account."

Does Opera Neon keep or train on your data?

It keeps it for a year and doesn't train on it. The privacy statement says your input and output are "stored, fully encrypted, on Opera's servers, and will be automatically deleted after 365 days." Opera's FAQ says the year exists so you can "return to and continue chats and agentic workflows," and that "Opera never trains AI models on your data."

Opera also says your data "is not used for advertising or personalization purposes." Its terms say Opera "does not claim ownership" of what you put in or get out and uses your input "solely for the purpose of providing these services."

The browser sends usage data of its own. The privacy statement lists Mixpanel and Firebase Analytics among Opera's technology providers, keeps feature usage data for up to three years, and keeps crash logs for up to five months. You can switch both off under Settings, then Privacy and security: turn off Help improve Opera by sending feature usage information and Automatically send crash reports to Opera.

Is it safe to let Neon Do use your logged-in accounts?

Your passwords and card numbers are safer than the pages they unlock. Opera's FAQ says that when you sign in to a site, "those credentials are not sent to any third-party servers (such as Opera's)," and that payment details "stay within your browser session and are never sent to Opera or any third-party servers." Neon Do still reads the pages behind those logins, and the FAQ says page content goes "to Opera's and OpenAI's servers."

Neon's terms are blunter than its FAQ. They say Neon "could also perform actions like accept cookie dialogs or consent to personal data processing on your behalf," that "it will not necessarily ask for your permission at every step," and that you shouldn't give its agents "information like your name, address, or credit card details." The privacy statement says Neon won't finish certain actions without your final approval and "will also block access to certain sensitive websites, like banking pages."

Websites may not know an agent is at work. Opera's CLI post says its command-line tool tells pages they're "being driven by automation" by default, but "Neon Do and Research operate without the flag regardless."

Opera's security post puts prompt injection first among the threats to agent browsers. It means text hidden in a page that the AI reads as orders, and our guide to prompt injection explains it. Brave reported a hidden-text attack on Neon on October 14, 2025 that made Neon read the user's email address from their Opera account page and send it to an outside server (Brave). Brave says Opera closed the report as "Not Applicable" on October 17. Opera's own account says its system rejected the report as a placeholder, that its team found it again on October 20, and that a fix was in production that night. The same security post says the risk of prompt injection "cannot be entirely reduced to zero."

What can other AI tools do through Neon?

As much as you allow, inside your real logged-in sessions. Since March 31, 2026, Neon can act as an MCP server, so tools like Claude Code, Lovable, and n8n can connect to it. Opera says that "any external AI system connecting through MCP can operate directly within those authenticated, real-time web sessions."

The defaults are narrow. Opera's post says three tools are on once you connect: Read tabs, Read page content, and Screenshot page. Clicking, typing, filling forms, and navigating are all "Write tools" that are off until you switch them on, and Read history is off too. The privacy statement says Opera doesn't store what passes to your tool and that the tool's own privacy policy applies.

Opera's CLI is the local path. It "runs on your machine," and its code is open source on GitHub. Our page on Perplexity Comet shows how much an agent with signed-in sessions can reach.

Who owns Opera?

Opera's own filing answers it. Opera Limited's annual report on Form 20-F, filed with the SEC on March 27, 2026, says Kunlun, "a Chinese public company listed on the Shenzhen Stock Exchange, indirectly owns approximately 68.0% of our outstanding ordinary shares." It says Opera's executive chairman, Yahui Zhou, is also Kunlun's controlling shareholder. The filing's risk factors add that because Kunlun is a China-based public company and Mr. Zhou is a Chinese national, "they are subject to the laws and regulatory environment of the PRC and actions by PRC authorities."

The same report says Opera's headquarters is in Oslo and its servers are in data centers "in the Netherlands, the United States, Canada, Nigeria, Iceland and Singapore." Neon's privacy statement names Opera Norway AS as the data controller, applies the GDPR, and lets you complain to Datatilsynet, Norway's data protection authority. Neon's terms are governed by Norwegian law, with disputes in the Oslo City Court. Our guide to where AI chat data is stored compares the controllers and countries behind 12 AI chat apps.

How to ask AI about what you read without a 365-day copy

Neon's year-long copy lives in your Opera account, next to your email address and your subscription record, and whatever Neon Make builds becomes a URL you can share. A Morse account holds less: your sign-in details, plan, balance, and a usage ledger with no message content. Your chat history is stored only in your browser, encrypted, and Morse's servers don't keep your messages or log what you write. Morse doesn't train AI on your content, sell your data, or show ads.

The models will look familiar. Neon's menu includes DeepSeek, GLM, Grok, GPT, Claude, and Gemini, and Morse bundles over 30 open and closed models in one account. Pick the mode by what you're asking about:

• Private, the default, for the page questions you'd send to Neon Chat. It runs open-weight models like DeepSeek, Kimi, and GLM, plus Grok. Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing, against Neon's 365 days on Opera's side and 30 days at OpenAI. Here's what zero data retention covers.
• Anonymous, for GPT, Claude, or Gemini, sent under Morse's account instead of yours. Before a message reaches the provider, our proxy strips the account, email address, and IP address it came from. The provider still reads the text, without knowing it's you, and its own policy may let it hold the message for a limited time to check for abuse. Before you paste a page in, delete the names on it.
• E2EE, for the signed-in pages you'd have let Neon Do read, like an order history or an insurance claim, pasted or typed in yourself. Your browser checks the secure enclave's hardware attestation itself, then encrypts the message so only that enclave can read it, and Morse's servers relay ciphertext they can't open. The guarantee rests on the enclave hardware and attestation working as designed. These chats use open-weight models and can't search the web, and they're part of Pro, which costs $19 a month. Our guide to end-to-end encrypted AI chat walks through the checks.

Signing out locks that history, and a copy of the browser's data can't be read on its own. Because your device holds the only copy of your history, clearing the browser or closing your Morse account deletes it, and Morse can't bring it back, so copy out anything you'll want later.

Send a question about the article in your other tab from the Morse home page as a guest, and choose "Keep this chat" if you want it saved, or sign up free without a card.