Technology
Risks of using AI at work

Risks of using AI at work

This survey helps assess key risks when using AI in work tasks: which risks employees consider most likely, what issues they have encountered in practice, which types of data and actions create uncertainty, and which measures (rules, tools, training, prompts) are seen as most helpful to reduce risks and improve safety.

Use template

The "Risks of using AI at work" template assesses how employees perceive and experience the downsides of using AI tools for work tasks. It combines perceived likelihood of specific risks with real incidents people have already faced, the data they refuse to share, and the safeguards they apply. Use it to build a realistic risk picture and prioritize the controls that matter.

What the “Risks of using AI at work” survey measures

The survey captures AI usage frequency, whether respondents have already hit a risk or issue, and which concrete situations they encountered — factual errors, wrong decisions driven by AI output, accidental sharing of internal data, hard-to-verify results, copyright/licensing exposure, inappropriate wording, or uncertainty about whether AI use was even allowed. A perceived-risk slider and a likelihood matrix rate threats such as confidential data leakage, sharing personal data, work-affecting errors, misread context, copyright violations, non-compliance and reputational risk. It also maps the data people deliberately keep out of prompts (customer and employee personal data, passwords/tokens, restricted files, financial metrics, contracts, internal code, plans and strategies), the risk-reduction actions they take (anonymize, use approved tools, verify facts, ask for rationale, avoid high-stakes tasks), and which measures would help most.

Who the “Risks of using AI at work” template is for

It fits security, risk, compliance and data-protection teams building an AI risk register, IT leaders defining approved tools and controls, and managers deciding where AI is safe to use. Organizations preparing an AI governance framework, or investigating incidents after early adoption, get evidence on both perceived and experienced risk in one survey.

How to adapt the template to your needs

Align the risk and data lists with your threat model — add sector-specific concerns (regulated data, trade secrets) or remove code-related items for non-technical teams. Add branching so people who report a real incident get a follow-up open question describing what happened, and route non-users straight past the safeguards block. You can turn the likelihood matrix into your own risk taxonomy, adjust the controls list to what you can technically enforce, or add a severity question alongside likelihood.

Questions and answer options

Question type: single choice.
Answer options:
— Daily
— Several times a week
— About once a week
— Less often
— I do not use AI
Question type: multiple choice.
Answer options:
— The output contained factual errors or made-up data
— The AI result led to an incorrect decision/action
— The prompt accidentally included internal data that should not have been shared
— It was difficult to verify the correctness of the output
— The AI used materials that could create copyright/licensing risk
— The output contained wording that was inappropriate for a work context
— It was unclear whether using AI for this task was allowed
Question type: single-answer matrix.
Answer options:
— Leakage of confidential information
— Sharing personal data with AI
— Errors/inaccuracies that affect work outcomes
— Misinterpretation of context or requirements
— Copyright/licensing violations when using outputs
— Non-compliance with internal rules/regulatory requirements
— Reputational risk due to inappropriate wording
Question type: multiple choice.
Answer options:
— Customers’ personal data
— Employees’ personal data
— Passwords, tokens, access keys
— Internal documents and files with restricted access
— Financial data and metrics not intended for distribution
— Information about customers/contracts/terms
— Code or code snippets from internal repositories
— Plans, strategies, non-public decisions
Question type: multiple choice.
Answer options:
— Remove/anonymize sensitive data before prompting
— Use only company-approved tools/accounts
— Verify key facts, numbers, and links before using the output
— Ask the AI to show steps/rationale for the result
— Compare the output with internal sources/documents
— Do not use AI for high-criticality tasks
— Save results only in approved locations
— Confirm AI use with a manager/security team when unsure
Question type: ranking.
Answer options:
— Short “do’s and don’ts” rules with examples
— Built-in prompts and warnings while entering a request
— A list of approved tools and use cases
— Prompt templates for safe common tasks
— A quick consultation channel (chat/form) for borderline cases
— Regular short training (10–15 minutes)
— Technical controls (data masking, blocking sensitive pastes)

Similar survey templates

Frequently asked questions

The likelihood matrix and risk slider capture what people fear, while the situations question records what has actually happened — errors, accidental data sharing, wrong decisions. Comparing the two shows whether your team over- or under-estimates real exposure and where perception and reality diverge.
It reveals both their risk instincts and where your policy is working. If sensitive items like passwords, customer data or internal code are widely avoided, self-protection is strong; if not, you've found a concrete gap to close with rules or technical controls.
It shows which safeguards are already habitual — anonymizing data, using approved tools, verifying facts, confirming with security — versus which are rare. Low adoption of a key control tells you exactly what to reinforce through training or built-in tooling.
Yes — 'the prompt accidentally included internal data that should not have been shared' is a distinct situation option, and confidential/personal data leakage appears separately in the likelihood matrix, so you can size this specific exposure on its own.