Mississippi Artificial Intelligence Network

AI Policy Supplement for Agriculture and Forestry

Decide who may share operational data, which AI recommendations need review, and who can authorize a change in the field.

Use alongside MAIN’s Business AI Policy and Guidance Template.

What this is
Adaptable planning guidance with sample clauses, a vendor checklist and a decision record.
Who it serves
Mississippi producers, forest landowners and managers, cooperatives, agricultural service businesses and their advisers.
Start here
Choose one proposed use. Resolve its data permissions, reviewer and operating boundary before approval.

01 · Scope

Choose the governing baseline

Use the Business AI Policy and Guidance Template for general employee use, HR, marketing, financial controls, confidentiality and procurement. Add this supplement where work involves farm or forest information, professional recommendations, field observations or connected equipment. Include AI embedded in existing software, not only chatbots. [1]

Crop, livestock, poultry, aquaculture and forestry operations can use the same decision structure while choosing different evidence and qualified reviewers. A mill or processing plant should also consult the Manufacturing AI Playbook. Public agencies and universities should begin with their own adopted policies and the relevant MAIN policy guide.

Official requirements
Applicable law, permits, binding contracts and adopted organizational policies remain controlling. Identify the actual activity and jurisdiction; a sector label does not determine applicability.
Voluntary guidance
NIST frameworks and Ag Data Transparent principles inform this supplement. They do not certify an operation or establish a blanket legal right to farm data. [2] [3]
MAIN recommendations
The review steps, proposed approval boundaries and sample clauses below are MAIN’s synthesis for local adaptation.

Boundary: this supplement supplies no pesticide, veterinary, engineering, drone-flight, harvesting or equipment-operating instructions. Obtain the qualified review and authorizations that the actual work requires.

Back to contents

02 · People

Assign authority before sharing or acting

Land ownership, equipment ownership, platform access and permission to reuse information are different questions. Check the lease, service agreement, data agreement and any other relevant terms. Where authority is unclear, resolve it before disclosure. The following roles can be combined in a small operation; record any need for an external reviewer. [3]

  • Accountable owner: approves the exact use and accepts responsibility for its limits.
  • Data contact: checks whose information is involved and documents permission to share it.
  • Qualified reviewer: checks outputs against appropriate field evidence, records or professional sources.
  • Operator and technical support: manage the approved configuration, safe operation and recovery.
  • Backup contact: handles concerns when the primary owner is unavailable.

Sample policy language to adapt

“[Operation Name] assigns [Owner] responsibility for [Approved Use]. Employees, contractors and advisers may use only the tools, data and permissions recorded for that use. [Reviewer] checks consequential outputs. [Stop Authority] may suspend AI reliance and initiate the approved fallback. A landowner, operator or vendor account alone does not establish authority to disclose another party’s information.”

Record next: one owner, a backup, the data-permission basis, the reviewer and the person authorized to stop the use.

Back to contents

03 · Data and contracts

Review what leaves the operation

Make a short inventory: field or tract maps, imagery, production history, animal or pond records, equipment telemetry and business information. Specify what the proposed tool actually needs. Removing names does not necessarily make a distinctive location, image or record anonymous. Assess what could still identify the operation or people. [4]

Ag Data Transparent’s 2024 principles address consent, provider relationships, AI training and data exit. Use them to ask concrete questions; their ownership position is an industry principle, not a determination of your legal rights. [3]

Sample policy language to adapt

“Before [Data Category] enters [Tool], [Data Contact] records permission, purpose, approved recipients and retention. [Approver] reviews the vendor agreement and available settings. Unresolved permission or material data-use terms prevent that data use from proceeding. Approval covers the listed account, feature and integrations; it does not authorize other uploads or connections.”

Vendor checklist: ask for evidence

For each item, record the answer, the contract clause or documentation, the reviewer, and a decision: acceptable, needs change or not applicable, with a reason. A sales demonstration is not contractual evidence.

Questions for the proposed tool and data use
AskEvidence to retain
Who receives the data, including connected providers and advisers?The agreement and a data-flow description for the selected features.
May inputs, outputs or telemetry train models or support other uses?The applicable clause and the settings that implement the agreed limits.
What can users view, change or send? How is access ended?A role/permission description and the process for removing staff or contractor access.
Can needed records be retrieved after a dispute, outage or exit?An export example, supported format and applicable availability/deletion terms.
What happens after a material model, feature or contract change?The change-notice process and the operation’s reapproval decision.
Who reports a failure, supports recovery and communicates an incident?Named support routes, relevant commitments and the local response owner.

Basis: Ag Data Transparent for agricultural data relationships; NIST AI RMF and Generative AI Profile for third-party risk, documentation and data controls. This checklist is MAIN’s adaptation, not a required federal form. [2] [3] [4]

Back to contents

04 · Evidence

Check recommendations against the work

Set the evaluation before relying on the output. Identify the location, season, crop or species, equipment and intended decision. A result from another setting may not establish suitability here. Generative AI can produce plausible but incorrect information. [2] [4]

  1. Choose the reference: current observations, approved records, equipment documentation or appropriate professional guidance.
  2. Test representative conditions: include missing, stale and conflicting inputs, and conditions in which a missed problem would matter.
  3. Set the decision threshold: specify acceptable errors, reviewer workload and conditions that stop reliance. Do not invent one accuracy target for every task.
  4. Keep the comparison: retain sources, corrections, reviewer reasoning and the outcome. Reassess after material changes.

Sample policy language to adapt

“[Reviewer] checks [Output] against [Reference Evidence] before it informs [Decision]. The approved use excludes [Out-of-Scope Conditions]. If evidence is missing, contradictory or outside the tested setting, the reviewer pauses reliance and routes the issue to [Qualified Person]. Vendor confidence scores do not replace this review.”

Official requirement example: EPA explains that pesticide labels are legally enforceable. AI output is not a substitute for the product’s labeling or the determination of applicable use requirements. Refer pesticide-related decisions to a qualified person using the current labeling. [8]

Back to contents

05 · Operating authority

Separate advice from permission to act

Record whether the system drafts text, reads operational information, recommends a change or can actually execute one. Equipment, irrigation, feeding, ventilation and other physical controls need a different review from a newsletter draft. NIST’s OT guidance addresses systems interacting with the physical environment and emphasizes safety and reliability. [5]

Sample policy language to adapt

“Approval to summarize or recommend does not authorize AI to change a setpoint, activate equipment, apply an input or alter a safety function. Consequential actions require the qualified person and approval recorded for that use. Automated control, if proposed, requires separate technical, safety and applicable professional review, defined operating limits, tested fallback arrangements and documented authorization. This supplement grants none of those permissions.”

For proposed connected use, have the qualified technical owner distinguish read access from control access, review remote/vendor permissions, and assess changes before deployment. Include embedded AI features and AI-created scripts. Retain the Business template’s software and change-review provisions. [1] [5]

Back to contents

06 · Continuity

Prepare for interruption and error

Decide what happens if the tool, connection, vendor or reviewer becomes unavailable during critical work. Identify the trusted records and approved process needed to continue. A fallback may be a qualified manual process or a controlled pause; it must be feasible for the actual operation. [5] [7]

  • Use individual approved accounts and multifactor authentication where offered. Protect and test needed backups. NIST provides practical small-business safeguards. [6]
  • Have the technical owner plan equipment/network changes and updates with compatibility and safety in view; do not apply an ordinary office-computer update routine blindly to equipment. [5]
  • Define reportable events: unauthorized disclosure or action, materially wrong advice, unexplained behavior, loss of required records, or a failure affecting people, animals, property or the environment.
  • Route suspected incidents to the named owner and appropriate responders. Preserve relevant evidence safely; have counsel or the responsible specialist assess notification duties. Do not invent a universal reporting deadline. [7]

Sample policy language to adapt

“Personnel report [Incident Types] through [Contact/Backup]. [Stop Authority] suspends reliance when [Stop Conditions] occur and follows [Approved Fallback]. Technical containment must respect established safety procedures. [Responsible Owner] documents the event, corrections and evidence supporting restoration before the use resumes.”

Back to contents

07 · Adoption

Adopt only what you can maintain

Walk through one real proposed use with the people who own the work. Resolve the relevant contract, technical and professional questions; adapt the clauses through your authorized policy process. Train staff and contractors to explain the allowed data, required review and stop conditions in their own words.

Review before a materially different season, location, crop/species, model, data connection, vendor term or operating purpose. Also set a calendar review date and revisit incidents. These are recommended review triggers, not a statutory review schedule. General AI learning is available through MAIN courses; completion does not authorize field operations. [2]

Blank decision record

Copy this record into your own system. Keep sensitive details in the approved location, not in a public page or an unapproved AI tool. An incomplete record is a request for review, not an approval.

Use and boundary
[Operation / proposed task / intended users / location and season / explicitly excluded work]
People
[Accountable owner / backup / data contact / qualified reviewer / technical support / stop authority]
Tool and permissions
[Provider / product and version / features / approved account and integrations / read, recommend or act]
Data and terms
[Allowed and excluded data / permission basis / vendor evidence / training, sharing, retention and exit decisions]
Requirements to confirm
[Applicable law, permits, labels, contracts and adopted policies / qualified owner / source and date checked / unresolved items]
Evidence and limits
[Reference evidence / tested conditions / errors and corrections / acceptance criteria / reviewer time / limitations]
Stop and recover
[Stop conditions / contact route / feasible fallback / evidence to retain / restoration authority]
Decision
[Approve defined use / revise and resubmit / hold / decline] · [Rationale / approver / date / conditions / review date and triggers]

Ready for a local decision?

  • People can explain exactly what the system may do.
  • Data permission and material vendor terms are resolved.
  • A qualified reviewer can check the result with available evidence.
  • Any consequential action has the required separate authorization.
  • The fallback works for the operation, and staff know when to use it.
  • The decision, unresolved limitations and next review are recorded.

Back to contents

Source notes

Sources and status

Sources checked October 25, 2026. The clauses and fictional examples are MAIN’s proposed adaptation; no source endorses or approves this supplement. Recheck the relevant original source when adopting or revising local rules.

  1. MAIN Business AI Policy and Guidance Template. Companion planning resource; general governance and AI-created software provisions. Not adopted policy.
  2. NIST AI Risk Management Framework 1.0, January 26, 2023. Voluntary framework; roles, context, evaluation and monitoring. NIST reports that revision is underway.
  3. Ag Data Transparent Core Principles, updated 2024. Voluntary agricultural industry principles. Inform data-contract questions; not a statute, legal ownership determination or forestry-specific standard.
  4. NIST AI 600-1: Generative AI Profile, July 26, 2024. Final voluntary companion guidance on generated information, data and third-party risks. Not validation of an agronomic or forestry model.
  5. NIST SP 800-82 Rev. 3: Guide to OT Security, September 2023. Final technical guidance; use qualified adaptation. Rev. 4 is an initial public draft, not a final replacement, as checked October 25, 2026.
  6. NIST Cybersecurity Basics, updated August 26, 2026. Practical voluntary safeguards for small businesses; equipment-specific compatibility still needs review.
  7. NIST SP 800-61 Rev. 3: Incident Response Recommendations, April 3, 2025. Final guidance; response planning and recovery. Does not set universal notification deadlines or supply animal-health/emergency operating procedures.
  8. EPA: Introduction to Pesticide Labels, updated June 2, 2026. Official explanation of enforceable labeling. Confirm the actual product, use, applicable exceptions and state requirements with the qualified reviewer.

For supporting career context, see AI Across Mississippi Careers. For organizational help, contact MAIN. Mississippi Artificial Intelligence Network is Mississippi’s statewide AI initiative.

Back to contents

Search MAIN

Courses, guides, news, and information.