- By Prateek Levi
- Mon, 17 Nov 2025 12:45 AM (IST)
- Source:JND
Perplexity is rolling out a new wave of improvements to its Comet Assistant, focusing heavily on safety, user oversight, and better visibility into how the AI operates when handling sensitive tasks. Announced on Friday, November 14, the upgrades are aimed at giving users clearer authority whenever the AI browser agent steps into what the company calls “high-stakes decisions”.
One of the major changes is that Comet Assistant will no longer proceed with actions like logging into websites or completing online purchases without pausing to ask for explicit confirmation. According to Perplexity, “The Assistant is designed to operate under the same principles a human assistant would, making some minor decisions without interrupting you while knowing to ask permission for more important decisions before acting on your behalf.”
ALSO READ: POCO F8 Pro And F8 Ultra Set For Global Debut: Expected Release Window Revealed
The update also expands what users can see as the AI works. The Assistant sidecar now shows a step-by-step breakdown of what the browser agent is doing and why, giving users a much clearer understanding of its decision-making process. Perplexity has added more direct controls as well, including buttons to halt the Assistant or guide it more precisely. The company says all these additions reflect its foundational principles of transparency, user control, and good judgement.
“Every improvement to the Comet Assistant advances these principles. It now shows you exactly what it’s doing, asks how you want it to act, and demonstrates sound judgement before completing sensitive actions,” the startup said.
The Perplexity Search bar—its omnibar—has also been refined. Users can now choose to browse manually, allow Comet Assistant to browse just once, or let it automatically step in whenever it detects an opportunity to help. These changes arrive as AI startups push aggressively into AI-assisted browsing, a space that has raised its own set of privacy and security worries.
Researchers have warned that agentic browsers like Comet and OpenAI’s ChatGPT Atlas are vulnerable to indirect prompt injection, which could expose personal data ranging from emails to banking credentials. OpenAI itself has previously acknowledged this type of threat, saying, “Besides simply making mistakes when acting on your behalf, agents are susceptible to hidden malicious instructions, which may be hidden in places such as a webpage or email with the intention that the instructions override the ChatGPT agent’s intended behaviour. This could lead to stealing data from sites you’re logged into or taking actions you didn’t intend.”
Perplexity, however, continues to emphasise that its approach keeps users firmly at the centre of control. “By prioritising visibility, adapting to your preferences, and requesting permission before sensitive actions, the Comet Assistant works at the speed of autonomy while keeping you firmly in control,” the company noted.
The startup recently enabled the AI agent to switch between browser tabs for tasks such as filling spreadsheets and conducting structured data entry. It also introduced architectural improvements to help the agent interact more intelligently with complex websites. With its valuation climbing and competition heating up, the company says it will introduce a new AI assistant in the coming days.