Opera Neon

Browser

Opera

Product overview

Name of Agent: Opera Neon
Short description of agent: "Opera Neon is a browser built to act. It works on your tasks, interprets the web, manages tabs, acts when you command, and helps you move faster." (link, archived)
Date of release: 30/09/2025 (link, archived)
Advertised use: "opening tabs, conducting research, finding the best prices, assessing security, whatever you need." (link, archived) and "filling in forms, placing orders, replying to emails, or tidying up files (link, archived)"
Monetisation/Usage price: (20, archived)
Who is using it?: end-users
Category: Browser

Company & accountability

Developer: Opera
Name of legal entity: Opera Norway AS
Place of legal incorporation: Oslo, Norway (link, archived)
For profit company?: Yes
Parent company?: Opera
Governance documents analysis: ToS (link, archived)
AI safety/trust framework: None found
Compliance with existing standards: Norwegian Transparency Act (link, archived)

Technical capabilities & system architecture

Model specifications: "Neon runs on Opera’s AI engine, which is model‑agnostic. We use different models depending on the task. Today, our primary providers are OpenAI and Google’s general‑purpose models. We also use specialized models for image generation, speech‑to‑text, text‑to‑speech, and so on. The agent picks the right model for the job—if you ask it to generate an image, it calls an image model; if you ask it to research, it uses the model we’ve found best for browsing and synthesis. This mix can evolve over time." (link, archived)
Documention: No specific docs, FAQ (here, archived).
Observation space: User input, web pages (link, archived)
Action space: Responses back to user, actions on the web (link, archived).
Memory architecture: "Neon uses AI memory that learns high‑level preferences over time—where you live, basic likes/dislikes, and so on. It’s early days; personalization will get better as we go." (link, archived)
User interface and interaction design: Agent works in the browser in a sandboxed browser environment that's similar to a user starting with a blank Opera browser. Sidebar shows agent's action trace and has chatbox for the user to prompt agent
User roles: Operator (directing the agent to complete tasks), Executor (user can take additional actions themselves after the agent has done working). No thumb up/down button to give feedback in "Do" mode, but is there in chat mode
Component accessibility: Closed source

Autonomy & control

Autonomy level and planning depth: L3: agent asks user for feedback and plan confirmation before taking more complex actions. Has an explicit "needs user action" UI to set the agent off at some points in the trajectory
User approval requirements for different decision types: User responds to requests for clarification from the agent and needs to approve plan before agent starts working. Also needs to click a play button on some steps that require user input ([source, archived)]
Execution monitoring, traces, and transparency: Visible CoT and action trace documenting all activity
Emergency stop and shut down mechanisms and user control: User can pause/stop the agent at any time
Usage monitoring and statistics and patterns: User can observe agent activity in the browser and CoT, unclear if other mechanisms exist

Ecosystem interaction

Identify to humans?: None found
Identifies technically?: - User-agent databases show Neon identifying itself through regular browser UA strings that do not make the browser distinguishable (link, archived)
Interoperability standards and integrations: None
Web conduct: None found

Safety, evaluation & impact

Technical guardrails and safety measures: Technical safeguards (link, archived) are built into the browser, prompt-injection safeguards by analyzing prompts for potentially malicious characteristics, human-in-the-loop approach, pausing for user interaction for sensitive actions like completing transactions or downloading files, blacklists high-risk pages (e.g., banking sites) to prevent AI agents from accessing/acting on them
Sandboxing and containment approaches: Yes (link, archived), cloud-hosted virtual machine (link, archived) for agent work “beyond the browser” used for Neon Make but Neon Do runs locally within your browser and directly interacts with web pages
What types of risks were evaluated?: None found
(Internal) safety evaluations and results: None found
Third-party testing, audits, and red-teaming: None found
Benchmark performance and demonstrated capabilities: None found
Bug bounty programmes and vulnerability disclosure: Yes (link, archived)
Any known incidents?: Prompt injection flaw in Opera Neon (link, archived) reported through BugCrowd
← Back to 2025 Index