As a baseline, do not paste passwords, authentication secrets, private keys, unreleased commercial information, unnecessary personal data, privileged legal material or customer records into an unapproved AI service. The exact prohibited list should come from your organisation’s policy and risk assessment.
Separate usefulness from permission
The fact that an AI tool can help with a task does not automatically mean every piece of workplace information belongs in it. Separate usefulness from permission: what data is involved, which tool is approved and who remains responsible for the output?
Credentials and secrets can create direct security exposure if disclosed outside their intended system. That fact should guide the decision before convenience does, particularly when the task touches sensitive information or a consequential outcome.
What matters in practice
Credentials and secrets can create direct security exposure if disclosed outside their intended system.
Data classification comes first. Public information, internal working material, personal data, credentials and commercially sensitive information should not all be treated as if they carry the same risk.
Personal data remains subject to data-protection requirements when an AI tool is used to process it.
Tool choice matters because business and consumer configurations can have different contractual, retention and administrative controls. Employees should know which service is approved before moving real work into it.
Data minimisation means using only the personal data needed for the defined purpose, not everything that happens to be available.
Human review remains important when the output affects customers, staff, money, compliance or safety. Productivity gains are useful only if responsibility for the final decision stays clear.
Where workplace AI gets risky
Workplace AI decisions depend on context. The same prompt may be harmless with public information and unacceptable with credentials, personal data or commercially sensitive material, so blanket yes-or-no rules are usually too crude.
- An approved enterprise AI environment may have different contractual and technical controls from a public consumer service.
- A generic list cannot replace sector-specific confidentiality, professional or regulatory duties.
What could change the answer?
Review the decision when the tool, contract, data type or business use changes. Moving from drafting public copy to handling customer records or making consequential decisions is a meaningful change in risk, even if the software name stays the same.
What you can do now
Start with low-risk uses and approved tools. Add sensitive data, automation or consequential decisions only when the organisation has the controls and accountability to support them.
- Keep passwords, keys and recovery codes out of prompts.
- Replace real customer or employee details with placeholders where possible.
- Ask for approval before using sensitive internal documents in a new AI service.
A four-question reality check
When a claim sounds dramatic, work through these four questions before deciding how much weight to give it.
What can the system demonstrably do?
What tools, credentials and networks can it actually reach?
Can it keep operating after errors or intervention?
What real-world outcome follows if safeguards fail?
Bottom line
The safest way to adopt workplace AI is in stages: start with low-risk tasks, learn where it helps, then add stronger controls before giving it more sensitive data or more consequential work.
As a baseline, do not paste passwords, authentication secrets, private keys, unreleased commercial information, unnecessary personal data, privileged legal material or customer records into an unapproved AI service. The exact prohibited list should come from your organisation’s policy and risk assessment.
Sources & further reading
These links are selected for the claims and methods discussed above. Primary or official sources are preferred where available.
Two useful questions
What is the biggest caveat?
An approved enterprise AI environment may have different contractual and technical controls from a public consumer service. A generic list cannot replace sector-specific confidentiality, professional or regulatory duties.
What is the most useful next step?
Keep passwords, keys and recovery codes out of prompts. Replace real customer or employee details with placeholders where possible.