In the horde of developing the best AI tools and in relentless pursuit of using these tools for our everyday work, are we compromising our security?

Researchers and multiple audits of Perplexity AI’s Android app found that it consists of several security flaws that can be easily exploited and lead to various kinds of cyberthreats, including data theft, account takeovers, and even impersonation attacks from cybercriminals. One of the flaws also allows users to access Perplexity’s API for free, causing huge revenue loss to the company itself.

Perplexity is a known and popular tech company offering one of the best AI assistants today. It was released nearly at the same time as ChatGPT and ever since its launch, it has been widely recognized as one of the most accurate chatbot offering well researched responses.

Raghunandan J, head of R&D at Appknox which discovered the vulnerabilities in Perplexity AI’s mobile app, says, “organizations today are wanting to come out with their applications and their updates faster, thereby ignoring their security”.

Let us dive deeper and understand what these vulnerabilities and security flaws the app presents.

Top Vulnerabilities in Perplexity

Researchers at Appknox discovered several vulnerabilities in DeepSeek this February, and now the same is also found in Perplexity. It includes:

• Unsecured network configuration – This can lead to an increase in network-based attacks
• No SSL validation or certificate pinning – It can be a reason for impersonation attacks
• Weak root detection – It can escalate privileges
• Susceptibility to the StrandHogg vulnerability – This vulnerability can be used to carry out phishing attacks or identity theft
• Exposure to Janus vulnerability – Attackers can modify APKs and inject malware because of this
• Tapjacking attacks – It can be used to manipulate UI interactions and trick users

Apart from these, there are five additional critical security issues in this AI assistant android app that can expose users to a variety of risks.

Newly Identified Security Issues in Perplexity AI

Below are some of the newly identified vulnerabilities in Perplexity’s app:

1. Hardcoded secrets found

   Attack type – Exposing the user’s credentials

   Risk – Critical

   Impact – Attackers can extract API keys by decompiling the app, which can lead to unauthorized access to backend services, potential data leaks, and compromise the security of systems, such as a SaaS provider breach in 2019.

   Fixes – Perplexity AI can securely store keys using environment variables or encrypted vaults. Using secure token management systems to rotate API keys regularly can also be helpful. A robust access control policy must be implemented to prevent unauthorized access.

2. CORS Misconfigurations (Wildcard Origin)

   Attack type – Cross-site request forgery (CSRF) and data theft

   Risk level – High

   Impact – This vulnerability lets any website make requests to the app’s backend, from where malicious sites can perform unauthorized actions on behalf of users and can even steal sensitive data.

   Fixes – restrict CORS origins to trusted domains. Implement proper authentication and token validation. Regular review of Perplexity’s API security configurations.

3. SSL Pinning Not Implemented

   Attack type – Man-in-the-middle (MITM) attack

   Risk level – Critical

   Impact – This vulnerability can lead to interception attacks where cybercriminals can spoof secure connections and steal user data.

   Fixes – It requires implementation of SSL pinning that can verify server certificates along with enforcing HTTPS with HSTS. Conducting periodic security testing for network vulnerabilities can significantly minimize the threats as well.

4. Bytecode Not Obfuscated

   Attack type – Reverse engineering and code tampering

   Risk level – Very High

   Impact – App logic is easily reverse-engineered, which leads to exposure of critical code paths and vulnerabilities. This was once observed with a popular ride-hailing app in 2020, where the attackers created a fraudulent version of the app and bypassed payment systems, which led to substantial financial losses to the company.

   Fixes - Use code obfuscation tools that are readily available in the market to prevent reverse engineering.

5. ADB Detection and Missing Developer Options Protections

   Attack type – Debugging exploitation

   Risk level – High

   Impact – Attackers can exploit this vulnerability in the app to easily manipulate it in a controlled environment using debugging tools.

   Fixes – Implement runtime checks to detect debugging tools and enforce security policies to restrict debugging in production environments.

Finally, why should you care?

If you are a Perplexity app user on your phone, then your personal data might be at risk because of these security issues.

Here are some risks to be aware of:

• Your sensitive information can be exposed
• You may never know when your personal and confidential data is stolen
• Malicious actors can spy on your activity
• Your phone’s security can be compromised

Therefore, it is recommended to uninstall the app immediately and use only trusted apps, especially when you are sharing sensitive information.

Key Takeaways

AI security debt is worsening with Perplexity compounding Deepseek’s mistakes. Though Deepseek has its own security flaws, critical to use data and devices, such as unsecured network configurations, the overall trend highlights how vulnerable these new AI tools are. It indicates the growing AI security gap where developers are focusing more on app development, ignoring security considerations, leaving users vulnerable. Therefore, users must use only trusted and most secure apps and avoid sharing personal information unless absolutely required.