Versatile AI Risk Assessment

Glossary Threats and weaknesses

Prompt Injection

Manipulated input or planted content redirects a language model.

As of: September 2026 · Glossary with 28 terms

What is Prompt Injection?

Prompt injection refers to attacks in which manipulated input or planted content (for example in documents, websites or emails) redirects the behaviour of a language model. The model then follows the attacker’s instructions instead of the operator’s rules, for example to exfiltrate confidential data or to bypass safeguards.

Effective controls combine input and output filtering, a strict separation of instructions and data, minimal permissions for connected tools, and testing with known attack patterns. The residual risks remain subject to documentation because no filter protects completely.

OWASP LLM01:2025MITRE ATLASEU AI Act Art. 15

Threats using this term (10)

The first 8 of 10 entries in the threat catalogue whose text uses the term. Every link leads to the full threat page.

Prompt Injection – Direct Attackers write instructions straight into their input so the AI system ignores its original rules and follows their commands instead. Prompt Injection – Indirect Malicious instructions hide inside external content such as documents, web pages, or emails. Meta Prompt Extraction Attackers get the AI system to reveal its hidden system prompt, configuration, or internal instructions. Sensitive Information Disclosure An AI system discloses confidential information without anyone intending it: trade secrets, personal data, credentials or internal … Exfiltration from ML Application Attackers or insiders copy data, model weights or configurations straight out of a production AI application without authorisation, … Insecure Tool Design Many AI applications act on their own through connected tools, for example sending emails or querying databases. Excessive Agency An AI system is granted more permissions or freedom to act than its task requires. Agentic AI / Autonomous Agents AI agents plan multi-step tasks on their own and carry them out with tools such as email, databases, or code execution.

More terms from the subject area Threats and weaknesses.

Hallucination (confabulation) Plausible sounding but factually wrong output, inherent to the technology. Data Poisoning Manipulated training or reference data steers a model wrong on purpose. Model theft and model extraction Theft of model weights or reconstruction of the model through the interface.

From the term into the substance

The link leads to the place on the website where the term becomes practical; the overview shows every term in the glossary by subject area.

To the prompt attack group in the catalogue All terms in the glossary

Cite this term

For reports, policies or internal documents; the link leads directly to this term page.

“Prompt Injection”. Versatile AI Risk Assessment, glossary of AI risk analysis, as of September 2026.
https://www.versatile-ai-risk-assessment.com/en/wissensbasis/glossary/prompt-injection/

← Back to the glossary