What is prompt injection?
Prompt injection is an attack where hidden instructions inside content an AI system reads, such as a document, a ticket, an email or a web page, trick it into ignoring its own rules and doing something it should not, like revealing confidential data or triggering an unauthorised action. The attacker needs no access to your systems, only the ability to put text where your AI will read it.
The name comes from older injection attacks such as SQL injection, where untrusted input is treated as a command instead of as data. Direct prompt injection is when a user types the malicious instruction themselves. Indirect prompt injection is the bigger risk in everyday work, because the instruction sits hidden in a file, a comment, a calendar invite or a web page that an AI assistant reads on your behalf, and a language model cannot reliably tell your instructions apart from text that only looks like instructions.
In work and project management this makes prompt injection a shadow AI problem rather than a developer one. Any AI agent that reads tasks, comments and attachments can be steered by anyone able to write into them, so the defence is architectural rather than a better prompt:
- Scope what the AI can reach. Role-based permissions mean an injected instruction inherits limited rights instead of full access.
- Keep a human approval step. Any action that changes data, sends a message or grants access should be reviewable and reversible.
- Log every AI action with its source. An audit trail lets you trace a suspicious output back to the content that caused it.
- Control where the model runs. Dedicated or on-premises deployment keeps untrusted external content away from your project data by default.
Related content
What are the risks of shadow AI?
The risks of shadow AI cluster into four areas: data leakage (confidential code, customer records, or trade secrets pasted into public tools that may store or train on that input), compliance violations (data handling that was never reviewed, breaching rules like GDPR or HIPAA), unreliable outputs (unverified AI results feeding real decisions), and expanded security exposure (unvetted apps that bypass controls and leave no audit trail).
What is shadow AI?
Shadow AI refers to employees using AI tools like ChatGPT, Copilot, or other generative AI services for work without the knowledge, approval, or oversight of their organization's IT or security teams. The main risks are data leakage, compliance violations, and loss of governance: sensitive company or customer data pasted into unsanctioned tools can be exposed, retained, or used to train external models.
