Polar AI Browser Launches With a New Model for Workplace Automation

Polar is introducing an AI browser designed to work inside web applications, not just display them. Announced on July 29, 2026, the new browser comes with a $5.7 million seed round led by Madrona. Built as a macOS-only, Chromium-based browser, it uses AI agents to complete multistep tasks within websites where a user is already signed in, a strategy that could reduce manual handoffs and the need for separate integrations for each service. At the same time, it raises new questions for IT teams around permissions, data handling, and oversight.

According to the article, CEO Kevin Jiang, who previously worked on Perplexity's Comet browser, said Polar targets recurring knowledge-work tasks rather than occasional consumer activities. The article notes that Polar can click, type, and navigate websites through existing browser sessions, and can schedule jobs using the current page or selected tabs for context. It claims users initiated more than 4.5 million web actions during seven months of testing, with some tasks running over 15 hours, but these figures have not been independently verified.

Polar states that users can observe an agent, take control at any time, and keep sensitive decisions in human hands. It also says guardrails are designed to prevent agents from independently completing high-risk actions, although detailed documentation of these controls remains limited. The company's privacy policy says browser history, bookmarks, cookies, local storage, and imported passwords generally stay on the device, but agent tasks may still send prompts, screenshots, page context, files, tool results, and conversation history to Polar’s systems and model providers. The article says the company claims its commercial provider agreements prohibit training on customer data and provide zero data retention, with limited exceptions.

The article highlights security concerns. Browser agents can encounter hostile instructions hidden in webpages, emails, or documents, a risk demonstrated by BioShocking attacks that tricked AI browsers into exposing sensitive data. OWASP classifies this as indirect prompt injection, which can divert an agent or trigger unauthorized actions. Approval controls also require testing, and the article cites a Claude for Chrome flaw where a rogue extension could trigger tasks in Gmail, Google Docs, and Google Calendar, with increased risk when unattended operation is enabled.

Before connecting Polar to production accounts, teams should verify: identity and offboarding with multifactor authentication and least-privilege access; approval gates defining which actions require confirmation; data handling documenting what leaves the device and where it is processed; audit and emergency controls ensuring logs and administrator stop functions exist; and injection defenses testing hostile instructions. The article suggests organizations start with synthetic data, nonproduction identities, and limited permissions, then expand access only after controls are confirmed and Polar provides sufficient documentation on sessions, approvals, logs, retention, and administration.