Perplexity Upgrades Comet Assistant With Stronger AI Safeguards
Perplexity has introduced major upgrades to Comet Assistant, focusing on safer automation and improved user control. The AI agent now pauses before completing sensitive actions such as website logins or online purchases.
Although the assistant can still handle low-risk tasks independently, it redirects important decisions to the user to ensure oversight. Moreover, these guardrails are designed to reduce risks linked to autonomous browsing and high-authority permissions.
Greater Transparency and Real-Time Monitoring
Users can now view the AI agent’s reasoning step-by-step inside a redesigned sidecar. Additionally, clear controls allow users to stop the assistant or give more guidance instantly. Because transparency is a priority, the assistant displays every action it takes and adapts to individual preferences over time.
The upgraded omnibar also offers a choice between manual browsing, single-session automated browsing, or fully automatic browsing when beneficial.
Competitive Push Amid Security Concerns
AI companies are racing to build AI-powered browsers, and this shift has amplified privacy and cybersecurity concerns. Consequently, researchers have warned that malicious hidden prompts could exploit browser agents to access confidential information. Furthermore, the rise of agentic browsing has sparked discussions about responsible design and safe deployment.
Perplexity continues developing Comet Assistant and recently added support for switching across tabs for spreadsheet and data entry workflows. It has also improved the agent’s architecture for better perception of complex websites while preparing to release a new AI assistant in the coming days.
