AI Confidentiality Toolkit: Acceptable Use Policies, Vendor Term Reviews, and Incident Playbooks

Summary. This toolkit supplies ready-to-adapt templates for protecting trade secrets and other confidential information when employees and contractors use generative AI tools. It includes an AI acceptable use policy, a data classification matrix that maps information tiers to approved tools, an approved-tools register, a vendor term review worksheet, confidentiality clauses for employee, contractor, and counterparty agreements, a prompt-leakage incident playbook with an incident report form, onboarding and annual employee certifications, an exit interview script, a legal hold addendum for AI logs, and a protective order provision on AI tool use. Each template states its purpose, owner, and timing and is followed by drafting commentary tied to the Defend Trade Secrets Act, the reasonable-measures case law, the NLRA, and the rules of civil procedure. A running example follows a flavor formulation company through each form. Bracketed placeholders mark every choice the drafter must make.

When a trade secret case reaches the reasonable-measures question, the owner's lawyer opens a binder. The forms in this toolkit are what should be in it: the documents that show a company identified its secrets, gave its people safe tools, told them the rules, made them sign, watched for mistakes, and fixed the ones it found. Each template is short enough to be used and specific enough to be proved.

The short answer

• Ten templates cover the program. Policy, matrix, register, vendor worksheet, clauses, incident playbook and report, certifications, exit script, hold addendum, and protective-order provision.
• Every form is evidence. Draft each one assuming it will be Exhibit 1 in a dispute over 18 U.S.C. § 1839(3)'s reasonable-measures element.
• Two legal guardrails run through all of them: the whistleblower immunity notice of 18 U.S.C. § 1833(b) and an NLRA savings clause protecting activity under 29 U.S.C. § 157.
• Adapt, do not paste. The placeholders are choices; a policy that names tools the company does not use, or tiers it never assigned, undermines itself.

How to use this toolkit

Each template states its Purpose, its Owner, and When to use it, then gives the form with [BRACKETED PLACEHOLDERS] and a short drafting commentary. Use one set of tier labels (T1 Restricted, T2 Confidential, T3 Internal, T4 Public) across every form so that the policy, the matrix, the DLP rules, and the incident report all mean the same thing.

The doctrine behind the forms is in Trade Secrets and Generative AI: Reasonable Measures, Prompt Leakage, and Vendor Terms; the sequence for building the program is in Building a Generative AI Use Policy That Protects Trade Secrets: A Practical Guide; and the item-by-item controls are in the Generative AI Trade Secret Protection Checklist: A Practical Checklist.

Running example. Kestrel Flavor Systems, Inc. is a 300-person New Jersey company that formulates flavors for beverage and snack makers. Its trade secrets are its flavor formulas (ingredient lists, ratios, and processing steps), its sensory-panel data, and its customer-specific reformulation briefs, many of which are customers' own confidential information. Kestrel's general counsel is Dana Whitcomb; its CISO is Arjun Mehta.

Template 1: AI acceptable use policy

Purpose: set the rules for using generative AI tools with company information. Owner: legal, with HR and security. When to use: at program launch, reviewed at least annually and whenever the approved-tools register changes materially.

[COMPANY NAME] GENERATIVE AI ACCEPTABLE USE POLICY
Effective date: [DATE]    Version: [X.X]    Owner: [TITLE]

1. PURPOSE
This Policy lets [COMPANY] personnel use generative AI tools productively
while protecting [COMPANY]'s trade secrets and confidential information,
information entrusted to us by customers and partners, personal data, and
privileged material.

2. SCOPE
This Policy applies to all employees, contractors, consultants, temporary
staff, and interns ("Personnel") who use any generative AI tool for
[COMPANY] work, on any device, including personal devices used for work.

3. DEFINITIONS
"Generative AI Tool" means any software or service that generates text,
code, images, audio, or other content in response to prompts, including
chat assistants, coding assistants, meeting assistants, browser extensions,
agents, and features embedded in other software.
"Approved Tool" means a Generative AI Tool listed in the Approved AI Tools
Register at [LOCATION], used through a [COMPANY]-provisioned account.
"Information Tiers" are T1 Restricted, T2 Confidential, T3 Internal, and
T4 Public, as defined in the Data Classification Matrix at [LOCATION].
"Connector" means any feature that gives a Generative AI Tool access to
email, files, calendars, code repositories, or other systems.
"Agent" means a Generative AI Tool configured to take actions, such as
sending messages, editing files, or calling other services.

4. CORE RULES
4.1 Use only Approved Tools, through [COMPANY]-provisioned accounts, for
    any [COMPANY] information other than T4 Public information.
4.2 Put information into an Approved Tool only if the Data Classification
    Matrix permits that tier in that tool.
4.3 Never put T1 Restricted information into any Generative AI Tool
    except [the self-hosted tool / the tool(s) listed for T1 use cases]
    for the use cases listed in the Matrix.
4.4 Never use a personal AI account for [COMPANY] information.
4.5 Do not enable Connectors, plug-ins, or Agents unless they are listed
    in the Register for your role.
4.6 Do not create public share links to conversations or publish custom
    assistants containing [COMPANY] information.
4.7 Do not put information belonging to a customer, supplier, or partner
    into any Generative AI Tool unless the Register shows that the
    relevant agreement permits it.

5. PROHIBITED USES (ALL TIERS)
5.1 Using any Generative AI Tool to process information obtained from a
    current or former employer, competitor, or other third party that you
    are not authorized to use.
5.2 Using Generative AI Tools to scrape, harvest, or extract information
    from others' systems or services contrary to their terms or technical
    restrictions, or to attempt to extract another service's prompts,
    model weights, or training data.
5.3 Attempting to bypass any technical control on Generative AI Tools.

6. OUTPUTS
6.1 Review outputs for accuracy before relying on them.
6.2 Treat any output that contains or is derived from T1 or T2
    information as information of the same tier.
6.3 Do not represent machine-generated material as solely human-authored
    where accuracy matters, and consult [LEGAL] before claiming copyright
    in material generated substantially by AI.

7. REQUESTS AND EXCEPTIONS
Request new tools, connectors, or exceptions through [PROCESS]. Only
[TITLE] may approve exceptions, in writing, for a stated period.

8. MONITORING
[COMPANY] monitors and logs use of Approved Tools and access to
Generative AI Tools from [COMPANY] systems and networks, as described in
[MONITORING NOTICE], to protect its information and comply with law.

9. REPORTING MISTAKES
If you put information into a Generative AI Tool in a way this Policy
does not permit, report it to [CONTACT] within [24 HOURS]. [COMPANY]
will not discipline Personnel solely for promptly self-reporting an
inadvertent mistake.

10. PROTECTED ACTIVITY
Nothing in this Policy prohibits or restricts any employee from
discussing or disclosing wages, hours, or other terms and conditions of
employment, or from engaging in other activity protected by Section 7 of
the National Labor Relations Act; from communicating with, filing a
charge or complaint with, or participating in an investigation by any
government agency; or from making disclosures protected by law.
[DTSA NOTICE: insert or cross-reference the notice in Template 5, Clause A.4.]

11. DISCIPLINE
Violations may result in discipline up to and including termination of
employment or engagement, consistent with [HANDBOOK SECTION].

12. REVIEW
[OWNER] will review this Policy at least annually.

Drafting commentary. Section 4.1 is the heart of the policy: it moves employees from personal accounts to company-provisioned ones, which is where the controls and logs live. The point of 4.4 is the lesson of Yellowfin Yachts, Inc. v. Barker Boatworks, LLC, 898 F.3d 1279 (11th Cir. 2018): information left on an employee's personal devices without restriction did not support reasonable measures, and a personal AI account is a personal device in the cloud. Section 5.2 addresses improper means from the other direction; automated harvesting of a competitor's semi-public data was held capable of being improper means in Compulife Software Inc. v. Newman, 959 F.3d 1288 (11th Cir. 2020), and a company whose employees do it invites claims. Section 9's self-report protection matters because incidents nobody reports cannot be contained. Section 10 is the savings clause; under the Board's Stericycle, Inc., 372 NLRB No. 113 (2023), standard, a rule that could reasonably be read to restrict Section 7 activity is presumptively unlawful under 29 U.S.C. § 158(a)(1). A savings clause does not cure a rule that is unlawful on its face, so the definitions must also be precise; the Board's composition changed in 2025, so check whether the standard has been revisited. For public companies, keep section 10's agency-communication language consistent with 17 C.F.R. § 240.21F-17.

Kestrel: Dana sets the self-report window at 24 hours, lists one enterprise assistant and one coding assistant in the Register, and leaves T1 with no permitted tool until a self-hosted pilot is approved.

Template 2: Data classification matrix for AI

Purpose: tell Personnel, on one page, which information may go into which tool for which purpose. Owner: legal and security. When to use: published with the policy; updated whenever a tool, tier, or overlay changes.

Drafting commentary. Calibrating precautions to value is the cost-benefit balance that Rockwell Graphic Systems v. DEV Industries, 925 F.2d 174 (7th Cir. 1991), describes; a matrix that is strict for T1 and permissive for T3 is evidence of that calibration. Keep the examples concrete and specific to the company. Health information requires a business associate agreement before disclosure to a vendor that handles it for a covered entity, under 45 C.F.R. Part 164. Export-controlled technical data, including items on the Commerce Control List in 15 C.F.R. Part 774, requires export counsel's review before it goes into any service with foreign servers or personnel.

Kestrel: customer reformulation briefs are T2 but flagged "third-party," so they may go into the enterprise assistant only for customers whose agreements are cleared in the Register.

Template 3: Approved AI tools register

Purpose: record each approved tool, its terms, and what it may receive. Owner: security, with legal sign-off. When to use: on each approval; reviewed quarterly.

Drafting commentary. The register links the policy to the contract. A court evaluating reasonable measures will want to see that the company knew which terms governed each tool, which is the "assurance of privacy" that distinguishes protected disclosure from release; the Supreme Court used that concept to define "confidential" information in Food Marketing Institute v. Argus Leader Media, 588 U.S. 427 (2019). Recording the agreement's version date matters because vendors change terms.

Template 4: Vendor term review worksheet

Purpose: evaluate whether an AI tool's governing terms support confidential use, and at which tier. Owner: legal, with procurement and security. When to use: before approval; on renewal; whenever the vendor announces a term change.

Drafting commentary. Disclosure to a vendor bound by confidentiality, for the owner's own purposes, is the kind of limited disclosure that does not destroy secrecy, as in Metallurgical Industries Inc. v. Fourtek, Inc., 790 F.2d 1195 (5th Cir. 1986). Disclosure under terms that let the recipient use the material looks more like the loss of the property interest described in Ruckelshaus v. Monsanto Co., 467 U.S. 986 (1984), where data submitted with notice of the government's use rights lost protection. Questions 3, 4, 6, and 9 do most of the work. Question 11 matters because vendors may be ordered to preserve content in litigation to which the customer is not a party, as conversation logs were in the 2025 news-publisher litigation against OpenAI. For unilateral changes (question 12), the FTC has warned that quietly expanding data use through new terms may violate 15 U.S.C. § 45, but negotiate a contractual right anyway. The commercial terms are covered in the AI Vendor Procurement and Governance Checklist: A Practical Checklist and the data-license side in the AI Licensing Toolkit: Data Provenance Records, License Terms, and Indemnity Clauses.

Kestrel: the worksheet for the enterprise assistant finds that the vendor's connector marketplace is governed by third-party terms, so Arjun disables it and the Register says so.

Template 5: Confidentiality clause updates for AI tools

Purpose: create the contractual duty that makes AI misuse a breach of a duty of secrecy. Owner: legal. When to use: template refresh, new hires and engagements, and contract renewals.

Clause A: Employee confidentiality agreement

A.1 AI Tools. Employee shall not input, upload, or otherwise disclose
Confidential Information to any generative artificial intelligence tool,
service, model, agent, or feature ("AI Tool") except an AI Tool that
[COMPANY] has approved for that category of information, accessed
through an account provisioned by [COMPANY], and used in accordance with
[COMPANY]'s Generative AI Acceptable Use Policy as amended from time to
time.

A.2 Personal Accounts. Employee shall not use any personal AI Tool
account for Confidential Information. If Employee discovers that
Confidential Information is held in any personal AI Tool account,
Employee shall promptly notify [CONTACT], and shall delete it when
directed and certify deletion in writing.

A.3 AI Work Product. Prompts, prompt libraries, outputs, datasets,
evaluation sets, fine-tuning data, model weights, and agent
configurations that Employee creates within the scope of employment
or using [COMPANY] resources are [COMPANY] property and, to the extent
not public, Confidential Information, and are assigned to [COMPANY]
under Section [INVENTION ASSIGNMENT SECTION].

A.4 Notice of Immunity. Pursuant to 18 U.S.C. § 1833(b), an individual
shall not be held criminally or civilly liable under any federal or
state trade secret law for the disclosure of a trade secret that (A) is
made (i) in confidence to a federal, state, or local government
official, either directly or indirectly, or to an attorney, and
(ii) solely for the purpose of reporting or investigating a suspected
violation of law; or (B) is made in a complaint or other document filed
in a lawsuit or other proceeding, if such filing is made under seal. An
individual who files a lawsuit for retaliation by an employer for
reporting a suspected violation of law may disclose the trade secret to
the individual's attorney and use the trade secret information in the
court proceeding, if the individual files any document containing the
trade secret under seal and does not disclose the trade secret, except
pursuant to court order.

A.5 Protected Activity. Nothing in this Agreement prohibits Employee
from discussing or disclosing wages, hours, or other terms and
conditions of employment, from engaging in activity protected by
Section 7 of the National Labor Relations Act, or from communicating
with any government agency.

A.6 Return at Separation. On separation, Employee shall identify every
AI Tool account Employee used for [COMPANY] work, delete Confidential
Information from any account not controlled by [COMPANY], and certify
deletion in the form provided.

Clause B: Contractor and consultant agreement (add to Clause A text)

B.1 Contractor represents that Schedule [X] lists every AI Tool that
Contractor or its personnel will use in performing the Services.
Contractor shall not process [COMPANY] T1 or T2 information in any AI
Tool other than those provisioned by [COMPANY]. Contractor shall flow
these obligations down to its personnel and permitted subcontractors
and is responsible for their compliance.

Clause C: Outbound NDA or commercial agreement (counterparty receives the company's information)

C.1 Recipient shall not input, upload, or otherwise disclose Discloser's
Confidential Information to any AI Tool unless (a) the AI Tool's
provider is contractually prohibited from using such information to
train or improve any model and from disclosing it to third parties,
(b) retention by the provider is limited to [30] days or less, or such
other period as Discloser approves in writing, and (c) access is
limited to Recipient's Representatives who need to know. Recipient
shall not use Discloser's Confidential Information to train, fine-tune,
or evaluate any model. Recipient is responsible for any disclosure by an
AI Tool provider used by Recipient as if made by Recipient.

Drafting commentary. A contractual duty of secrecy is what turns misuse into "breach of a duty to maintain secrecy," improper means under 18 U.S.C. § 1839(6), and misappropriation under § 1839(5). Without a confidentiality request, an owner risks the result in Fail-Safe, LLC v. A.O. Smith Corp., 674 F.3d 889 (7th Cir. 2012), where disclosure without confidentiality obligations defeated the reasonable-measures claim. Clause A.4 tracks the statutory notice; an employer that omits it in an agreement entered into or updated after May 11, 2016 cannot recover exemplary damages or attorney's fees under 18 U.S.C. § 1836(b)(3) against an employee who was not notified. Clause A.3's coverage of AI work product matters because copyright may not protect machine-generated outputs after Thaler v. Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025), leaving contract and trade secret law as the main protection. Clause C's "train, fine-tune, or evaluate" language closes a gap in older NDAs that restrict disclosure but not use. Broader NDA drafting is in Drafting Enforceable Non-Disclosure Agreements for Technology Transactions; assignment drafting is in Employee Invention Assignment Agreements: Drafting for Enforceability Across Jurisdictions.

Kestrel: Dana adds Clause C to the confidentiality terms in Kestrel's flavor development agreements, because co-packers and ingredient suppliers routinely receive partial formulas.

Template 6: Prompt-leakage incident playbook

Purpose: contain, assess, and document a disclosure of company or third-party information to an AI tool outside the policy. Owner: security (operations) and legal (decisions). When to use: on any self-report, DLP alert, or discovery of a disclosure.

Drafting commentary. Steps 3 and 4 come before containment for a reason. Deleting a conversation that later turns out to matter in litigation can support sanctions under Fed. R. Civ. P. 37(e), and deletion by a party's own employees produced adverse-inference proceedings in E.I. du Pont de Nemours & Co. v. Kolon Industries, Inc., 277 F.R.D. 286 (E.D. Va. 2011). Run HIGH-severity investigations at the direction of counsel to support privilege under Upjohn Co. v. United States, 449 U.S. 383 (1981). Step 7 matters because wide internet circulation can end secrecy, as recognized in Religious Technology Center v. Netcom On-Line Communication Services, Inc., 923 F. Supp. 1231 (N.D. Cal. 1995). Integrate this playbook with the company's general incident response plan; see the Data Breach and Incident Response Toolkit: From Detection to Notification and Cybersecurity Incident Response and IP Protection: Preventing Trade Secret Loss During Data Breaches.

Template 7: AI incident report form

Purpose: record what happened and what the company did, as evidence of reasonable measures. Owner: security, reviewed by legal. When to use: closing every incident, including low-severity ones.

AI INCIDENT REPORT                      Incident ID: [ID]
[PRIVILEGED AND CONFIDENTIAL: PREPARED AT THE DIRECTION OF COUNSEL]
(use the legend only where counsel directed the investigation)

1. Date reported / date of disclosure: [DATES]
2. Reported by: [SELF-REPORT / DLP ALERT / OTHER]
3. Tool, plan, and account type: [TOOL; PLAN; COMPANY OR PERSONAL]
4. Information involved (describe by category; do not reproduce):
   [DESCRIPTION]   Tier: [T1-T4]   Overlays: [LIST]
5. Third-party information involved: [YES/NO; COUNTERPARTY CLASS]
6. Containment actions and dates: [LIST]
7. Vendor confirmation received: [YES/NO; DATE; SUMMARY]
8. Public exposure found: [YES/NO; DETAILS]
9. Assessment: [e.g., "Partial fragment of Formula Family [X];
   no complete formula disclosed"]
10. Notifications made: [RECIPIENT; DATE; BASIS]
11. Remediation: [TRAINING; CONTROL CHANGE; OTHER]
12. Legal hold issued: [YES/NO; DATE]
13. Lessons and control changes: [TEXT]
Prepared by: [NAME, DATE]        Reviewed by: [COUNSEL, DATE]

Drafting commentary. Item 4 says "describe by category; do not reproduce" because an incident report that copies the secret becomes another copy to protect. Use the privilege legend only where counsel actually directed the investigation; a legend on routine operational reports is not a substitute for the facts that support privilege, and an overused legend invites challenge. A file of closed incidents, each handled promptly, is strong evidence that the company treated its information as secret.

Template 8: Employee AI certifications

Purpose: confirm that each person knows the rules and, at hiring, has not brought anyone else's secrets. Owner: HR. When to use: onboarding (Part A) and annually (Part B).

PART A: ONBOARDING CERTIFICATION
I, [NAME], certify that:
1. I have not brought to [COMPANY], and will not use or disclose in my
   work, any confidential information or trade secrets of any former
   employer or other third party, including information stored in any
   personal AI tool account, chat history, saved prompt, custom
   assistant, or file I uploaded to an AI tool.
2. I will not use any AI tool to reconstruct, from memory or otherwise,
   documents, code, formulas, or data belonging to a former employer.
3. I have received and read the Generative AI Acceptable Use Policy
   (version [X.X]) and will comply with it.
Signature: [ ]   Date: [ ]

PART B: ANNUAL CERTIFICATION
I, [NAME], certify that during the past year:
1. I have used only Approved Tools, through company-provisioned
   accounts, for [COMPANY] information other than public information.
2. I have not entered T1 Restricted information into any AI tool
   except as the Data Classification Matrix permits.
3. I have reported every mistake of which I am aware, or I list them
   here: [ ].
4. I do not hold [COMPANY] information in any personal AI tool account,
   or I list the accounts here for deletion: [ ].
5. I have completed the [YEAR] AI confidentiality training.
Nothing in this certification restricts my rights described in Section
10 of the Policy.
Signature: [ ]   Date: [ ]

Drafting commentary. Part A protects the company as a potential defendant. A new employer that acquires a trade secret knowing or having reason to know it came through improper means, or through someone who owed a duty of secrecy, can itself be liable under 18 U.S.C. § 1839(5); the certification, combined with training, is the company's evidence that it did not have reason to know. Item 2 of Part A addresses the AI-specific risk that a new hire will use a model to regenerate the former employer's materials, which the broad reading of "use" in Oakwood Laboratories LLC v. Thanoo, 999 F.3d 892 (3d Cir. 2021), would likely reach. Part B's self-disclosure items generate incident files rather than discipline. Training and acknowledgment records are among the measures courts credit; Learning Curve Toys, Inc. v. PlayWood Toys, Inc., 342 F.3d 714 (7th Cir. 2003), describes the contextual inquiry into whether the owner's precautions were reasonable.

Template 9: Exit interview script (AI accounts)

Purpose: close the AI channel at departure and create a record. Owner: HR, with escalation to legal. When to use: every departure; for employees with T1 access, schedule before the last day and coordinate with the forensic review in the departure checklist.

EXIT INTERVIEW: AI TOOLS SECTION
Interviewer: [NAME]   Departing employee: [NAME]   Date: [DATE]

1. "Which AI tools did you use for [COMPANY] work, including personal
   accounts, browser extensions, meeting assistants, and coding tools?"
   Record: [LIST]
2. "Do any of those accounts, histories, saved prompts, custom
   assistants, or uploaded files contain [COMPANY] information or
   customer information?"  Record: [YES/NO; DETAILS]
3. "Did you create any prompts, prompt libraries, datasets, or other AI
   materials for your work that are stored outside company systems?"
   Record: [YES/NO; DETAILS]
4. "Will you delete [COMPANY] information from any account we do not
   control today, in our presence or by [DATE], and sign the
   certification?"  Record: [RESPONSE]
5. "Is your new role likely to involve [DESCRIBE AREA OF T1 INFORMATION]?"
   Record: [RESPONSE]
6. Reminder delivered: continuing confidentiality obligations under
   [AGREEMENT]; return-of-property obligations; DTSA notice (Clause A.4).
7. Escalate to legal the same day if: any answer is evasive; the
   employee declines to delete or certify; logs show bulk uploads or
   summarization of large document sets in the last [90] days; or the
   employee is joining a competitor in the same field.
Departing employee certification of deletion: [ATTACH]

Drafting commentary. The script is a record, not a cross-examination. Ask, write down the answers, and escalate. Evidence of downloads and deletions before departure was central to the preliminary injunction in WeRide Corp. v. Huang, 379 F. Supp. 3d 834 (N.D. Cal. 2019), and the AI-era equivalent is a personal account holding summaries of company documents. Do not ask the employee to surrender personal account credentials; ask for deletion and certification, and, if necessary, pursue preservation through counsel. Any restriction on the new employment must rest on evidence of threatened misappropriation, not merely on what the person knows, under 18 U.S.C. § 1836(b)(3)(A). The full departure process is in the Employee Departure Trade Secret Protection Checklist: A Practical Checklist.

Kestrel: a senior flavorist leaving for a competitor answers question 1 by naming a personal chatbot account she used "for menus, mostly." Item 7 is triggered because she is joining a competitor; the forensic review finds no uploads from company devices, and the file records both facts.

Template 10: Legal hold addendum and protective order provision for AI

Purpose: make sure AI data is preserved when litigation is anticipated, and that the other side's AI use does not expose the company's secrets in discovery. Owner: litigation counsel. When to use: with the first legal hold notice; at the Rule 26(f) conference or when negotiating a protective order.

Part A: Legal hold addendum

AI TOOLS AND DATA. This hold covers information in generative AI tools,
including: conversation histories, prompts, outputs, uploaded files,
custom assistants, saved prompts, and share links in [APPROVED TOOLS]
and in any personal AI account you used for [COMPANY] work; logs held
by [COMPANY] security tools for those services; connector and agent
activity logs; and logs, prompts, and outputs in [SELF-HOSTED SYSTEMS].
Do not delete conversations or change retention settings. IT will
suspend automatic deletion for the accounts of [CUSTODIANS] in
[TOOLS] effective [DATE].

Part B: Protective order provision

Use of AI Tools With Protected Material. A Receiving Party may process
another Party's Protected Material with a generative AI tool only if
(a) the tool is deployed in an environment controlled by the Receiving
Party or its counsel or vendor, (b) the provider is contractually
prohibited from using the material to train or improve any model and
from disclosing it to third parties, (c) the provider retains no copy
beyond [30] days, or such shorter period as the Parties agree, and
(d) access is limited to persons authorized to view that designation
under this Order. No Party may input HIGHLY CONFIDENTIAL - ATTORNEYS'
EYES ONLY material into any AI tool without [the Producing Party's
written consent / the Court's leave]. At the conclusion of the
litigation, the obligation to return or destroy Protected Material
extends to copies held in AI tools, indexes, and embeddings.

Drafting commentary. Prompts, outputs, and logs are electronically stored information subject to Fed. R. Civ. P. 34, and retention settings that delete them after the duty to preserve arises create Rule 37(e) risk. Protective orders rest on Fed. R. Civ. P. 26(c), and in trade secret cases the court must also preserve confidentiality under 18 U.S.C. § 1835. Part B's last sentence addresses embeddings and indexes, which a return-or-destroy clause drafted for paper and email does not reach. Raise the provision early, because the other side's review vendor may already use AI tools. For the broader protective-order framework, see the Trade Secret Litigation Toolkit: Identification Statements, Two-Tier Protective Orders, and Forensic Protocols; for the lawyer's own AI use, ABA Formal Opinion 512 (July 2024) requires attention to confidentiality and, for self-learning tools, informed client consent before client information is input.

Putting the templates together: Kestrel's first year

Kestrel adopted Templates 1 through 5 in a single rollout: the policy, the matrix, a register with two tools, two completed vendor worksheets, and updated employee and development-agreement clauses. Every employee signed Template 8, Part A or B. In the first quarter, the DLP rule keyed to Kestrel's formula numbering scheme fired eleven times; nine were T2 customer briefs pasted into the enterprise assistant for customers not yet cleared in the Register, which led Dana to clear six more customer agreements and add a pop-up reminder. Two incidents involved a T1 formula fragment pasted by a junior chemist into the enterprise assistant. Template 6 ran to completion in both cases; each Template 7 report recorded deletion, the vendor's written confirmation that the content was not used for training, and retraining. When the flavorist left for a competitor, Template 9 produced a clean record. If Kestrel ever has to prove reasonable measures, it will not be describing intentions. It will be producing forms.

Practical takeaways

• Use one tier vocabulary across every form.
• Tie the policy to a register that names the exact plan and agreement version for each tool.
• Put the DTSA notice and an NLRA savings clause in every form that governs confidential information.
• Capture and hold before you contain, and describe secrets by category in incident reports.
• Ask about AI accounts at hiring and at departure, and write down the answers.
• Address AI tools and embeddings in legal holds and protective orders.

This document is general information, not legal advice for any particular situation; the law, vendor terms, and technology change, and the facts matter.

Related Documents

Related Articles

• Trade Secrets and Generative AI: Reasonable Measures, Prompt Leakage, and Vendor Terms: the doctrine behind each template.
• Licensing Data and Models for Artificial Intelligence: Training Rights, Outputs, and Indemnities: output ownership and data rights that interact with Template 4.
• Copyright in Software Interfaces After Google v. Oracle: APIs, Compatibility, and Reverse Engineering: the copyright limits behind the policy's prohibited uses.

Related Guides

• Building a Generative AI Use Policy That Protects Trade Secrets: A Practical Guide: when and how each template is used.
• Trade Secret Misappropriation Litigation Under the Defend Trade Secrets Act: A Practical Guide: where the forms become exhibits.
• Preparing and Filing a Provisional Patent Application: A Practical Guide: the alternative to secrecy for inventions that will be disclosed anyway.

Related Checklists

• Generative AI Trade Secret Protection Checklist: A Practical Checklist: the tasks these forms document.
• Employee Departure Trade Secret Protection Checklist: A Practical Checklist: the departure process around Template 9.
• Biometric Data Compliance Checklist: A Practical Checklist: additional rules when AI tools process biometric data.

Related Toolkits

• Trade Secret Protection Toolkit: Building and Defending a Confidentiality Program: the broader confidentiality program forms.
• AI Governance Toolkit: Inventory, Risk Classification, and Controls: inventory and risk tools for AI systems.
• Clean Room Toolkit: Protocols, Team Certifications, and Specification Records: certifications and records for independent development.