AI Security: Questions and Answers
Prompt injection, safe use of AI chatbots and agents at work, AI policies and AI frameworks such as the NIST AI RMF, ISO 42001 and the EU AI Act.
What is prompt injection?
Prompt injection is an attack where text supplied to an AI model, directly by a user or hidden in a web page, email or document the model reads, overrides the instructions the developer intended. It is listed first in the OWASP Top 10 for LLM Applications.
Full answer and next stepsIs it safe to put company data into ChatGPT or other AI chatbots?
Only if your organization has approved the tool and its data terms for that kind of data. Consumer AI accounts may keep conversations and, depending on settings and terms, use them to improve models. Business plans often offer stronger data controls, but you need to check the provider’s current terms.
Full answer and next stepsWhat is an AI acceptable use policy and what should it cover?
An AI acceptable use policy tells staff which AI tools they may use, for what, and with which data. It should cover approved tools, prohibited data, human review of AI output, disclosure to customers where relevant, who approves new tools, and how to report problems.
Full answer and next stepsHow do I secure an AI chatbot on my website?
Give the chatbot access only to the data and actions it truly needs, keep secrets out of the system prompt, filter and validate its output before it reaches other systems, rate-limit usage, log conversations for abuse review, and require human approval for anything that changes data or spends money.
Full answer and next stepsWhat is the OWASP Top 10 for LLM Applications?
It is an OWASP project that lists the most important security risks for applications built on large language models, including prompt injection, sensitive information disclosure, supply chain risks, data and model poisoning, improper output handling and excessive agency.
Full answer and next stepsWhat is data poisoning in AI?
Data poisoning is when an attacker manipulates the data used to train, fine-tune or ground an AI model, so the model learns harmful behavior, hidden backdoors or wrong answers. It also applies to retrieval systems that pull documents into a model’s context at run time.
Full answer and next stepsWhat is shadow AI and why is it a risk?
Shadow AI is the use of AI tools by staff without the organization’s knowledge or approval. The risk is that company or customer data ends up in services nobody has reviewed for security, privacy or contract terms, and that AI output is used in decisions without checks.
Full answer and next stepsWhat is the NIST AI Risk Management Framework (AI RMF)?
The NIST AI RMF is voluntary guidance for managing risks from AI systems throughout their life cycle. It is organized into four functions, Govern, Map, Measure and Manage, and NIST also publishes a Generative AI Profile that applies it to generative AI risks.
Full answer and next stepsWhat is ISO/IEC 42001?
ISO/IEC 42001 is an international standard that specifies requirements for an AI management system: the policies, roles, risk assessments and controls an organization uses to develop or use AI responsibly. Organizations can be certified against it, much as with ISO/IEC 27001 for information security.
Full answer and next stepsWhat is the EU AI Act and does it apply to my business?
The EU AI Act is a European Union regulation that sets rules for AI systems based on risk: some practices are prohibited, high-risk systems face strict requirements, some systems carry transparency duties, and most others have few obligations. It can apply to organizations outside the EU that place AI systems on the EU market or whose AI output is used in the EU.
Full answer and next stepsAre AI agents safe to use at work?
They can be, with guardrails. AI agents take actions such as sending email, editing files or calling APIs, so the main risks are excessive permissions and prompt injection. Give agents the least access needed, log their actions, and require human approval for anything irreversible.
Full answer and next stepsWhat is AI red teaming?
AI red teaming is structured, authorized testing of an AI system by people trying to make it fail: bypassing its safety rules, extracting sensitive data, triggering harmful output or abusing its tools. The findings feed fixes and ongoing monitoring before and after launch.
Full answer and next stepsWhat is the difference between AI security and cloud security?
Cloud security protects digital infrastructure, including servers, networks, storage, and cloud access controls. AI security focuses on protecting machine learning models, training data, prompts, and autonomous agent workflows. While cloud security secures the environment where models and applications run, AI security defends against threats unique to machine learning, such as prompt injection and data poisoning. You need both to protect modern AI workloads.
Full answer and next stepsCan you create your own AI agents?
Yes, you can create AI agents by connecting large language models to external tools, APIs, and data sources. These systems use model reasoning to execute multi-step workflows autonomously. Anyone with basic software development skills or access to agentic orchestration platforms can build them. However, giving autonomous software the ability to take real-world actions introduces significant operational and cybersecurity risks that require strict guardrails.
Full answer and next stepsGo deeper
Step-by-step articles on this subject are in the AI Security guides.