OWASP Top 10 for Large Language Model(LLM) Applications & Generative AI

GenAI relies heavily on LLM to process prompts from users.
The OWASP® Foundation has identified its top 10 potential threats present when using LLM.
Each item in the list is tagged with the OWASP Top 10 code beginning with LLM01 until LLM10.

The following briefly details outbound sensitive data threats identified by the OWASP Top 10 that are protected by IWS:

• LLM01:2025 Prompt Injection

  Manipulating LLMs via crafted inputs can lead to unauthorized access, data breaches, and compromised decision-making




• LLM02:2025 Sensitive Information Disclosure

  Failure to protect against disclosure of sensitive information in LLM outputs can result in legal consequences or a loss of competitive advantage.




• LLM05:2025 Improper Output Handling

  Neglecting to validate LLM outputs may lead to downstream security exploits, including code execution that compromises systems and exposes data.

• LLM06:2025 Excessive Agency

  Granting LLMs unchecked autonomy to take action can lead to unintended consequences, jeopardizing reliability, privacy, and trust.




• LLM07:2025 System Prompt Leakage

  Exploiting system prompts or instructions can lead to direct leakage of sensitive data by the LLM.

Threats Targeting GenAI Deployment Using LLMs

While using third-party or open-source GenAI LLMs, every Business Entity (BE) will need to manage the attendant supply chain and third-party risks.
For example, malicious open-source models or open-source models with poor security controls may introduce vulnerabilities or backdoors, which could also result in data leaks.

Data Leakage from GenAI Deployment by Employees

BEs that allow employees to use publicly accessible GenAI tools (e.g. ChatGPT and Gemini) could be subject to potential common data leak scenarios:

• Employees submit or upload sensitive data while using those tools.
• BEs could also be exposed to data leakage risks through unauthorised insider actions and improper data handling when using the GenAI solutions.

Data Leaks from Prompt Injection Attacks Against GenAI Solution(s)

In a prompt injection attack, the following are commonly observed scenarios:

• The threat actor provides a malicious input to lead the GenAI model into revealing sensitive information in the generated response.
• Instances of jailbreak attacks, where the threat actor uses specially crafted prompts to bypass the security controls and safeguards implemented to make the GenAI solution disclose sensitive information.

Consequences of Data Leakage from GenAI Deployment

Where such data leaks involve confidential information and intellectual property, the BE could end up facing:

• Legal consequences
• Regulatory consequences
• Reputational consequences

General Countermeasures Recommended for DLP

Current commonly deployed DLP measures are:

• implementations of purpose-built firewalls for GenAI models that analyse user inputs to detect any attempts to extract data or exploit the GenAI solutions.
• DLP controls to check for any sensitive data provided in the prompts, as well as the responses generated by their GenAI solutions.

DLP Outbound Security Gap

• GenAI firewalls and countermeasures alone are currently not fully able to prevent outbound data breaches should malicious compromises occur.
• IWS provides a crucial last line of defence against compromised outbound sensitive data from GenAI countermeasures failure.

Infotect's IWS GenAI DLP Capabilities

Infotect's patented innovations effectively detects outbound anomalies and provide 24 x 7 protection against:

• Leakages of personal, medical and financial records.
• Server configuration information.
• Malicious or compromised content in public websites and prevent it from being spread to unknowing site visitors.

IWS protection for DLP countermeasures: