Quick answer: A staffing agency AI policy should do six things: name the AI tools recruiters may use, ban personal AI accounts for agency work, control who can connect AI to the ATS or CRM through MCP, set rules for candidate and client data, keep a human on every hiring decision, and cut AI access the day someone leaves. A free template covering all six is below.
Why now: about half of workers already use AI tools their employer has not approved, and about half have connected AI to work systems without IT sign off. An agency without a policy has one anyway, written by its recruiters.
Who this is for: agency owners, ops managers and anyone asked to “write something up” before the team connects AI to the database.
Why does a staffing agency need its own AI policy?
Because your recruiters are already using AI, approved or not. A BlackFog survey of 2,000 UK and US employees found 49% use AI tools their employer has not sanctioned, and 51% have connected or integrated AI tools with other work systems without IT approval, according to BlackFog’s research. That second number is the one to worry about: it describes exactly what an MCP connection to your ATS is.
Leaders are not setting a better example. TrustedTech’s survey of 2,001 workers found senior decision makers are more than twice as likely to use unapproved AI tools as the people they manage, 65% against 31%, per TrustedTech’s report. A policy only works if the owner follows it too.
Staffing firms carry extra risk. American Staffing Association general counsel Brittany Sakata has pointed out that staffing firms are in the business of hiring people and their candidate databases can be tens of thousands deep, in a Jackson Lewis interview. More records means more exposure when something goes wrong.
What should an agency AI policy cover?
| Section | What it settles | Why it matters |
|---|---|---|
| Approved tools | Which AI products and plans recruiters may use | Business plans keep data out of model training and give you admin controls |
| Personal accounts | No agency data in personal AI accounts | Personal accounts leave with the recruiter |
| MCP and integrations | Who may connect AI to the ATS, CRM, inbox or files | A connection gives AI the same reach as the recruiter, at machine speed |
| Data rules | What candidate and client data may go into AI | Privacy notices, client contracts and sensitive fields |
| Human review | Where a person must decide | AI hiring laws and plain fairness |
| Offboarding and audit | How access ends and what gets logged | Departing recruiters are the biggest data risk |
The ASA’s own guidance points the same way. One member on its staffing technology taskforce described requiring documented use cases, defined human in the loop checkpoints and clear escalation paths for any AI capability, in ASA’s AI feature.
The free AI policy template for staffing agencies
Copy this into your handbook and edit the bracketed parts. It is a starting point, not legal advice; have your employment counsel review it before you roll it out.
[Agency name] AI Acceptable Use Policy
Effective [date]. Owner: [name, title]. Reviewed every six months.
1. Purpose and scope. This policy covers every employee, contractor and owner who uses AI tools for [Agency name] work, on any device. “AI tools” means chat assistants, agents, AI features inside our software, browser extensions and note takers.
2. Approved tools. You may use only the AI tools on the approved list kept by [owner]: [for example, Claude Team, ChatGPT Business, Microsoft 365 Copilot, AI features inside our ATS]. Ask [owner] to add a tool. Free and personal plans of any AI product are not approved for agency work.
3. Personal accounts. Do not put candidate, client, placement, pay or pipeline information into a personal AI account, even if you pay for it yourself. Use your company login only.
4. Connecting AI to agency systems (MCP and integrations). Only [owner or named admin] may connect an AI tool to our ATS, CRM, email, calendar, file storage or phone system, whether through MCP, a plugin, an API key or a browser extension. Every connection is listed in the connection register with its purpose, the account it uses and its access level. New connections start read only. Write access is added one action at a time, after [owner] approves it in writing.
5. Candidate and client data. Use the minimum data the task needs. Never put into any AI tool: Social Security or national ID numbers, bank details, medical or disability information, background check results, immigration documents, or anything a client has marked confidential. Follow our candidate privacy notice and each client’s contract.
6. Human decisions. AI may draft, summarize, search and suggest. A person decides who is submitted, rejected, interviewed, offered or placed, and checks AI output for accuracy before it reaches a candidate or client. Do not use AI to score or rank candidates for a decision without [owner]’s approval and any notice the law requires.
7. Files and links from outside. Resumes, emails and web pages can contain hidden instructions aimed at AI. Do not let an AI tool take actions (send, update, delete) based on content it read from an outside file without your review.
8. Exports and bulk requests. Do not ask AI to export, list or summarize large parts of the database (for example, every contact at every client) unless your manager asked for it. AI activity in our systems is logged and reviewed.
9. Leaving the agency. On your last day, [owner] removes your access to every approved AI tool and every connection you used, and reviews recent AI activity. Agency data in AI tools, chats, projects and memories belongs to [Agency name].
10. Reporting. If you think agency data went into an unapproved tool, or an AI tool did something unexpected, tell [owner] the same day. Reporting quickly is never a disciplinary matter; hiding it is.
Acknowledgement. I have read and will follow this policy. Name, date, signature.
Why ban personal AI accounts?
Two reasons. First, control: a personal account is outside your admin settings, your logs and your offboarding. When the recruiter leaves, their chat history, saved projects and any connections they built leave with them. Second, data terms: business plans from the major AI vendors generally keep your data out of model training by default, while consumer plans depend on each user’s settings. Our candidate data guide walks through the terms.
Some staffing firms already work this way. Procom’s chief operations officer told the ASA that after finding staff with their own ChatGPT or Claude licenses, the firm now allows only tools the organization defines and secures on its own network, per the same ASA article.
How should the policy handle MCP connections to the ATS?
As the most sensitive item in it. An MCP connection lets an AI assistant read and sometimes change your live database with the signed in user’s permissions. That makes it powerful and fast, which is the point, and also the risk. We cover what the platforms offer in agencies and MCP access, and Bullhorn’s third party options in our Bullhorn MCP guide.
Three rules do most of the work:
- One owner makes connections. Recruiters use them; they do not create them.
- Read only first. Add write actions one at a time, in writing.
- Permissions before AI. Tighten ATS roles before anything is connected, because the AI inherits them. Our guide to protecting your database lists the settings.
Which laws should the policy point to?
It depends where you recruit. The ASA’s legal sponsor Staffing GC advises staffing firms to inventory each AI tool, map it to the places it is used, schedule any legally required audits, build candidate notification workflows and standardize record retention, in ASA’s legal column. The same piece makes a simple rule worth copying: hold AI tools to the same legal standards as your human staff.
For the specific laws, from New York City’s bias audit rule to GDPR, see our AI and candidate data guide.
How do you roll the policy out?
- Find out what is already in use. Ask the team, without blame, which AI tools and extensions they use today.
- Pick the approved stack. Choose one assistant on a business plan. Our assistant comparison covers plans and admin controls.
- Set up the connection register. A simple sheet: tool, connection, account used, access level, approved by, date.
- Train for thirty minutes. Walk through sections 3 to 8 with real examples from your desk.
- Get signatures and review twice a year. AI products change monthly; your approved list should too.
Quick answers
Do small agencies need an AI policy?
Yes. A two person desk with a candidate database has the same data risks as a large firm, and the template above takes an hour to adapt.
Should we ban AI altogether?
Bans tend to push use underground. Surveys show many workers keep using AI even when it is not approved, so an approved, controlled option works better.
Can recruiters use ChatGPT on their phones?
Only through the company account on your approved plan, and never with candidate data in a personal app.
Who should own the policy?
One named person, usually the owner or ops lead, who keeps the approved list and the connection register.
Does the policy need to mention MCP by name?
Name it alongside plugins, API keys and extensions, so nobody can argue a connection did not count.
What about resumes with hidden AI instructions?
Section 7 covers it. See our guide to resume prompt injection for examples.
The bottom line
Your recruiters will use AI with or without a policy. Write down which tools are approved, keep agency data out of personal accounts, let one person make every database connection, start read only, keep a human on hiring decisions and close access the day someone leaves. For every guide in this series, see MCP for staffing agencies.
Sources: BlackFog, TrustedTech, Jackson Lewis, ASA AI Acceleration, ASA legal column
Free newsletter
Get the Work Insiders weekly
One email a week: the recruiting and HR tech news that mattered, the tools we just added to the directory, and one practical guide. No spam, and you can unsubscribe in one click.