Perplexity AI & How to Stay Safe with AI Tools

The Bug That Made Headlines

In August 2025, security researchers at Brave discovered a serious vulnerability in Perplexity's Comet browser—an AI agent that can browse the web and perform tasks on your behalf. The flaw was indirect prompt injection.

Here's how the attack worked:

1. A user visits a seemingly innocent Reddit thread or website.
2. Hidden malicious instructions are embedded in the content (white text on white background, HTML comments, or social media posts).
3. When the user asks Comet to "summarize this page", the AI can't distinguish between legitimate content and hidden commands.
4. Result: The AI could access login credentials, navigate to sensitive accounts, and send data to attackers—all without the user knowing.

The scary part?
This happened through normal AI usage—just asking for a page summary.

A Fellow’s Perspective: Encouragement, Transparency & Swift Action

As a Perplexity AI Business Fellow, I’ve seen firsthand how quickly Perplexity responds to security issues. When this issue was uncovered, Perplexity acted with urgency and transparency, patching the vulnerability and clearly communicating with the community.

At my recent CreateHER Fest workshop, I stressed two essentials:

• AI tools can massively boost your research, productivity, and creativity.
  Make Perplexity your favorite “coworker”—use it for research, project management, and technical brainstorming.
• Practical security must be part of everyone’s AI toolkit.
  Double-check AI suggestions, and never paste confidential data into prompts—smart security habits keep you and your data safe.

Perplexity’s rapid, responsible response is a model for the industry—speed, openness, and user safety are non-negotiable for modern AI tools.

What This Means for Using AI Tools

This isn't just about Perplexity. As AI agents become more powerful—able to browse, write emails, manage accounts—the security risks multiply. The vulnerability highlights a challenge:
AI tools can be manipulated through everyday content.

Real-world implications:

• Banking & healthcare sites accessed through compromised AI
• Email accounts potentially exposed during research tasks
• Corporate data at risk when AI processes infected documents
• Personal information leaked through routine AI interactions

The good news: Perplexity has patched this specific issue, but broader security challenges remain across the AI ecosystem.

How to Use AI Tools Safely (Without Giving Up Productivity)

🛡️ Smart Browsing with AI

• Don't use AI agents on sensitive, logged-in sites (banking, healthcare, work accounts).
• Log out of critical accounts before using AI browsing tools.
• Use separate browser profiles for AI tools vs. personal/work browsing.
• Be skeptical of AI suggestions to visit specific websites or take actions.

🔐 Data Protection Strategies

• Avoid sharing sensitive info in prompts (SSNs, passwords, confidential work data).
• Use anonymized examples when possible ("a company in my industry" instead of naming your employer).
• Check privacy settings regularly.
• Consider enterprise AI tools for work tasks (stronger security controls).

⚠️ Red Flags to Watch For

• AI suggesting unexpected actions (sending emails, accessing accounts)
• Unusual requests to visit websites or download files
• AI responses that seem off-topic or contain strange instructions
• Repeated prompts to take actions you didn’t request

📋 Verification Best Practices

• Double-check AI-generated information before acting.
• Verify sources independently.
• Review permissions before giving AI tools account access.
• Monitor account activity for unauthorized actions.

The Bigger Picture: AI Security is Everyone's Responsibility

As more women enter AI and emerging tech, security isn’t optional—it’s essential.

• If you’re building AI products:
  Consider security from day one.
• If you’re using AI for work:
  Advocate for secure policies.
• If you’re learning AI:
  Include security and ethics training.

Moving Forward: Smart AI Usage

The goal isn’t to avoid AI tools—they’re too valuable. Instead, use them thoughtfully:

• ✅ Use AI for research, writing, and analysis (with data precautions)
• ✅ Leverage AI for learning new skills and exploring tech
• ✅ Experiment with AI tools in safe, controlled environments
• ❌ Don’t use AI agents for sensitive account management
• ❌ Don’t share confidential data in AI prompts
• ❌ Don’t blindly trust AI suggestions for sensitive actions

Liked this article?
Subscribe to our newsletter for career stories, workshops, tips, tools, events, and to stay updated on AI security!