Acceptable Use of AI

What this policy covers, and who it applies to

This policy sets the standard for how AI is used across Beckons. It is a baseline that protects the Company, our guests, and the people we work with.

Who it applies to

• Everyone at BeckonsEmployees full-time and part-time, contractors, consultants, interns, temporary staff, and anyone else acting on behalf of the Company.
• Any device used for Company businessIncluding personal devices. The rules don't change because the laptop is yours.

What it applies to

• Generative AI toolsText, image, audio, and video generation tools of any kind.
• Coding assistantsDeveloper tools that write, complete, or review code.
• Analytics and data toolsAny tool that uses AI to process or interpret data.
• AI features in existing softwareThe AI features baked into tools you already use also count.

Approved tools and how to get access

You may only use AI tools on the Beckons Approved AI Tools list for Company business. Anything else is off-limits, regardless of how convenient or capable it appears.

Currently approved

• Anthropic ClaudeVia a Company-provisioned enterprise account.
• Microsoft CopilotVia a Company-provisioned enterprise account.

Requesting a new tool

If you believe another tool would be useful, submit a written request to the CFO and the Systems and Projects Manager before using it. Your request must include:

• Tool name and vendorWho makes it, and what it's called.
• Intended use caseWhat you'd actually use it for.
• Data types involvedWhat kind of information you'd put into it.
• Confirmation of paid tierFree tiers are not approvable.
• Confirmation that Company data will not be used for model trainingThis is a hard requirement.
• Estimated monthly costSo Finance can plan.
• If a Connector is involvedA Connector is an integration that lets an AI tool access or act in another system, including via the Model Context Protocol. Where a Connector is part of the request, also include details of the connected system and the access the AI will be granted.

What you can and cannot put into AI

Most data breaches involving AI come from someone pasting too much into a prompt. The rule is simple, and the rule is non-negotiable.

What counts as Protected Information

Protected Information is any non-public, confidential, sensitive, or regulated information about Beckons or third parties. Examples include:

The rule

• Approval is data-specificProtected Information must not be entered into any AI tool unless that specific tool is approved for that data type and the right safeguards are in place. Approval for one purpose does not mean approval for all data.
• Before you paste, ask three questionsDo I actually need to include this to get a useful answer? Can I anonymise it (remove names, identifiers, account numbers)? Can I summarise the situation without the underlying data?
• Unapproved tools never receive Protected InformationNo exceptions, no workarounds, no quick favours.

Your responsibilities when using AI

AI is a tool. The responsibility for the work product, and for the consequences, stays with you.

• Outputs are not factsAI outputs can be inaccurate, incomplete, biased, or simply wrong. You remain responsible for everything you produce, regardless of how it was produced.
• Review before anything goes externalBefore AI-assisted content reaches anyone outside Beckons (communications, customer-facing material, regulatory submissions, legal documents) it must be reviewed and signed off by the right person.
• Stand behind your workYou must be able to explain and defend any decision or communication where AI played a role. If you cannot, do not send it.
• DisclosureWhere the use of AI must be disclosed for legal or contractual reasons, ensure that disclosure happens. Do not assume someone else will catch it.
• Where AI must not run unsupervisedAI must not be used to make or automate decisions in legal, regulatory, employment, privacy, financial, or customer-facing contexts without appropriate human oversight.
• Watch for biasAI tools can reflect biases from their training data. Actively evaluate outputs whenever AI is used in contexts that affect people, and never use AI in ways that lead to unlawful discrimination.

What to do if something goes wrong

Mistakes happen. The thing that turns a small mistake into a big one is delay. Speak up early.

• Report it immediatelyIf you suspect or become aware of a breach of this policy, report it to the Systems and Projects Manager. Do not wait to see whether it becomes a real problem.
• No retaliationBeckons will not tolerate retaliation against anyone who raises a concern in good faith. Speaking up is encouraged and protected.
• ConsequencesBreaches may result in disciplinary action up to and including termination, and may carry legal consequences. The response will reflect the nature, severity, and circumstances of the breach.
• MonitoringBeckons may monitor AI tool usage on Company systems and networks to verify compliance with this policy.

Quiz

Ten questions drawn from across the five modules. A pass requires seven correct. The Submit button activates once you've answered every question.

Your results