--- title: "AI Memory Retention and Deletion Policy: A Checklist." description: "Write an AI memory retention and deletion policy for sources, summaries, indexes, backups, and product memory. Set rules and test removal with a canary." canonical: "https://scalewithsearch.com/articles/ai-memory-retention-and-deletion-policy" date: "2026-08-20" modified: "2026-09-19" --- ## Site navigation - [Scale With Search](https://scalewithsearch.com/) - Real estate - Real estate - [Real estate](https://scalewithsearch.com/for/real-estate) - Work - Start here - [Send your brief](https://scalewithsearch.com/work#send-your-brief) - [Prepare your six-question brief](https://scalewithsearch.com/work#prepare-your-six-question-brief) - Build - [Site build, content library with SEO, signal desk](https://scalewithsearch.com/work) - For your business - Trades and home services - [Auto body and collision shops](https://scalewithsearch.com/for/auto-body-and-collision-shops) - [Foundation and home repair contractors](https://scalewithsearch.com/for/foundation-and-home-repair) - [Garage door and fencing contractors](https://scalewithsearch.com/for/garage-door-and-fencing-contractors) - [HVAC contractors](https://scalewithsearch.com/for/hvac-contractors) - [Janitorial and commercial cleaning companies](https://scalewithsearch.com/for/janitorial-and-commercial-cleaning) - [Locksmiths](https://scalewithsearch.com/for/locksmiths) - [Moving companies](https://scalewithsearch.com/for/moving-companies) - [Pest control companies](https://scalewithsearch.com/for/pest-control-companies) - [Plumbing and electrical contractors](https://scalewithsearch.com/for/plumbing-and-electrical-contractors) - [Restoration and water or fire damage companies](https://scalewithsearch.com/for/restoration-and-water-fire-damage) - [Roofing companies](https://scalewithsearch.com/for/roofing-companies) - [Towing companies](https://scalewithsearch.com/for/towing-companies) - [Tree services and landscaping companies](https://scalewithsearch.com/for/tree-services-and-landscaping) - [Solar installers](https://scalewithsearch.com/for/solar-installation) - [General contractors](https://scalewithsearch.com/for/general-contractors-and-construction) - [Paving, concrete, and flooring contractors](https://scalewithsearch.com/for/paving) - Practices and professional services - [Bookkeeping and tax practices](https://scalewithsearch.com/for/bookkeeping-and-tax-practices) - [Dental practices](https://scalewithsearch.com/for/dental-practices) - [Family and criminal defense law firms](https://scalewithsearch.com/for/family-and-criminal-defense-law-firms) - [Med spas and aesthetics practices](https://scalewithsearch.com/for/med-spas-and-aesthetics) - [Personal injury law firms](https://scalewithsearch.com/for/personal-injury-law-firms) - [Veterinary clinics](https://scalewithsearch.com/for/veterinary-clinics) - [Gyms and fitness studios](https://scalewithsearch.com/for/fitness) - [Therapy and outpatient health practices](https://scalewithsearch.com/for/therapy-and-outpatient-health) - [Medical billing companies](https://scalewithsearch.com/for/medical-billing) - [Insurance agencies](https://scalewithsearch.com/for/insurance-agencies) - [Financial advisors](https://scalewithsearch.com/for/financial-advisors) - [Property management companies](https://scalewithsearch.com/for/property-management) - [Recruiting and staffing agencies](https://scalewithsearch.com/for/recruiting-and-staffing) - [Architects and interior designers](https://scalewithsearch.com/for/architects-and-interior-designers) - [Logistics and supply chain companies](https://scalewithsearch.com/for/logistics-and-supply-chain) - Agencies, MSPs, and manufacturing - [IT and managed service providers](https://scalewithsearch.com/for/it-and-managed-service-providers) - [Machine shops and precision manufacturers](https://scalewithsearch.com/for/machine-shops-and-precision-manufacturing) - [Marketing agencies and freelancers](https://scalewithsearch.com/for/marketing-agencies-and-freelancers) - [SEO agencies and consultants](https://scalewithsearch.com/for/seo-agencies-and-consultants) - [Small manufacturers and fabricators](https://scalewithsearch.com/for/small-manufacturers-and-fabricators) - Restaurants, shops, studios, and nonprofits - [Restaurants and hospitality businesses](https://scalewithsearch.com/for/restaurants-and-hospitality) - [Retail stores and ecommerce sellers](https://scalewithsearch.com/for/retail-and-ecommerce) - [Photographers, event planners, and travel agents](https://scalewithsearch.com/for/photographers) - [Churches and nonprofits](https://scalewithsearch.com/for/churches-and-nonprofits) - [All industries](https://scalewithsearch.com/for/) - Learn - For your office - [Office job guides](https://scalewithsearch.com/guides/) - [Browser calculators](https://scalewithsearch.com/tools/) - Start here - [How it works](https://scalewithsearch.com/how-it-works) - [Free Starter Kit](https://scalewithsearch.com/kit/business-memory-starter-kit.zip) - [Synthetic specimen](https://scalewithsearch.com/specimen/working-session-specimen.zip) - Guides - [The Complete Guide to Business Memory for AI Agents](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide) - [The Complete Small-Business Guide to AI Agent Governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business) - [The Complete Guide to Leaving Vendor AI Memory](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide) - Articles by cluster - [Business memory](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide) - [Agent governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business) - [Migration and ownership](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide) - For machines - [llms.txt](https://scalewithsearch.com/llms.txt) - [llms-full.txt](https://scalewithsearch.com/llms-full.txt) - [Machine view](https://scalewithsearch.com/?view=machine) - Company - Evidence - [Proof](https://scalewithsearch.com/proof) - Company - [About](https://scalewithsearch.com/about) # AI Memory Retention and Deletion Policy for a Small Business. A client asks you to delete their information. You remove the conversation from the chat sidebar and confirm the request. Their details still appear in an embedding index, a weekly summary, an agent profile, and last night's recovery copy. The visible chat was one copy. The business system had four more. An AI memory retention policy must cover every place a record can be stored, derived, retrieved, backed up, or remembered by a provider. It should name why each record exists, who owns the decision, when deletion begins, and how the business proves completion. ## AI memory has more copies than the chat shows Start with an inventory. Do not start by choosing a blanket period such as ninety days. AI work can leave records in source files, uploaded files, chat history, saved preferences, project instructions, generated summaries, vector indexes, caches, logs, approval records, backups, and external connectors. A deletion request aimed at one interface will not necessarily reach all those stores. OpenAI distinguishes archiving a chat from deleting it. Its retention guidance says deleted chats are scheduled for permanent deletion within thirty days, subject to stated exceptions, while archived chats remain in the account. [OpenAI: Chat and file retention policies](https://help.openai.com/en/articles/8983778-chat-and-file-retention-policies-in-chatgpt) The distinction matters for any vendor. Hiding, archiving, exporting, expiring, and deleting are different operations. Your policy should use one term for each. ## Inventory raw, derived, indexed, and backed-up memory For every recurring job, trace one client fact from entry to final receipt. Ask where the original entered, which process transformed it, where retrieval copies were written, which services received it, and what recovery systems can restore it. Include model-side saved memory when the product provides it. This is easier when durable records live in [plain-text AI memory](/articles/plain-text-ai-memory). Readable files expose what exists and support ordinary search, review, and deletion. They do not remove the need to inspect indexes, vendor workspaces, or backups. Use one row per record class in `ai-memory-retention.csv`: ```csv record_class,system,purpose,authority,retention_trigger,delete_from,exception,proof_owner client_decision,decisions.md,deliver current work,contract,account_closed+24m,"live,index,backup",legal_hold,account_owner meeting_audio,object_store,make approved transcript,consent,transcript_accepted+30d,"live,cache,backup",active_dispute,record_owner embedding,vector_index,retrieve approved context,source_record,source_deleted,"index,snapshot",none,system_owner agent_summary,summaries/,prepare weekly brief,derived,source_superseded+30d,"live,index,backup",none,account_owner run_receipt,receipts/,prove execution,operations,job_retired+24m,"live,backup",legal_hold,system_owner ``` The `authority` field prevents a generated summary from acquiring the same status as a signed decision. The `delete_from` field forces the policy to name every known copy. ## Keep records for a named reason Every retained class needs a business, contractual, security, or legal reason. "The AI may need it later" is not a reason. A current instruction may support a recurring job. An approval receipt may prove who authorized an external effect. A correction may prevent a known failure. A raw recording may have no continuing purpose after an accepted transcript and dispute window. Proton's small-business retention guide recommends a schedule that defines the data category, retention period, reason, owner, and deletion method. [Proton: Data Retention Policy](https://proton.me/business/blog/data-retention-policy) Legal requirements vary by location, contract, and record type. This article is an operating design, not legal advice. Have qualified counsel set legal periods and hold rules. Encode the approved result in the register so the agent does not invent it. ## Set retention by record type One period for all AI data is easy to state and hard to defend. Source facts, signed decisions, draft outputs, access logs, embeddings, temporary caches, and backups serve different purposes. Assign each class its own trigger. A trigger is often clearer than a calendar date: contract end, accepted transcript, superseded decision, closed incident, or revoked consent. Derived records should not outlive their sources by accident. If a source is deleted, mark every summary, embedding, extracted field, and cached prompt that depends on it. Either rebuild the derived store without that source or delete the affected object. The policy should also explain what happens when an account remains active but a fact becomes wrong. A correction can supersede a fact immediately while the old record remains restricted as evidence. That is different from pretending the old record never existed. ## Separate facts, assumptions, preferences, and action logs Persistent memory becomes risky when the system cannot tell an approved fact from an inference. Classify memory at entry: - confirmed facts cite an authoritative source; - decisions name the person or record with authority; - preferences state scope and owner; - assumptions expire or enter review; - generated summaries remain derived; - action logs record what happened without becoming policy. Microsoft's security research describes risks from poisoned or misleading memory and recommends validating origin and integrity before use. [Microsoft Security: Guarding AI memory](https://www.microsoft.com/en-us/security/blog/2026/06/22/guarding-ai-memory/) Your [context-reading rules](/articles/what-context-should-an-agent-read) should exclude expired and unreviewed classes. Retention does not mean automatic eligibility for every future job. ## Delete through the whole dependency chain Run deletion as a bounded workflow. 1. Resolve the subject and approved request. 2. Find the source records and stable identifiers. 3. Find derived records that cite those identifiers. 4. Remove or quarantine live records. 5. Rebuild affected indexes and summaries. 6. Expire caches and saved sessions. 7. Apply the documented backup process. 8. Query each active store for the identifiers. 9. Write a deletion receipt. Backups need a specific rule. Immediate physical removal may conflict with how immutable recovery copies work. State whether the record becomes inaccessible, ages out on schedule, and will be deleted again if a recovery restores it. Do not claim erasure from a backup you cannot inspect. ## Record exceptions and legal holds A deletion exception must name its authority, scope, owner, start date, review date, and release condition. Do not let the agent create a legal hold. It may detect a matching approved hold and stop the deletion. A qualified owner decides whether the hold applies. Keep held data out of ordinary retrieval when possible. Retention for a dispute does not make the record useful context for a sales draft or service summary. This custody test is part of deciding whether you [own the AI memory](/articles/do-you-own-ai-memory). If the business cannot locate, restrict, export, and delete the record, ownership is incomplete. ## Test deletion with a canary record Create a synthetic client and a unique canary value, such as `RETENTION-CANARY-8427`. Put it through the real ingestion path. Confirm the canary appears in the source file, retrieval index, generated summary, and recovery inventory. Then execute the deletion workflow. Test these results: 1. The source lookup returns no active record. 2. Semantic and exact-match retrieval return no canary. 3. The current summary no longer contains the value. 4. A normal agent run cannot reproduce it from allowed sources. 5. The backup register shows the applicable expiry or restore-delete rule. 6. A deletion receipt lists every store checked. 7. A second reviewer can repeat the searches. 8. An unrelated client's records remain unchanged. Fail the policy if the team checks only the chat interface. Also fail it if the receipt claims a store was cleared without recording the query or provider result. ## Approval and stopping boundary An agent may inventory stores, prepare proposed retention rows, identify dependent records, and run a synthetic canary test in an approved environment. It stops before deleting client data, changing a legal period, releasing a hold, modifying backup policy, or asserting that a request is complete. Those actions require the named record owner and any required legal review. The executor also stops when identity is ambiguous, two authorities conflict, a store cannot be queried, or deletion would affect unrelated records. The final receipt must distinguish confirmed deletion, scheduled expiry, inaccessible store, and unresolved exception. ## Sources - [OpenAI: Chat and file retention policies](https://help.openai.com/en/articles/8983778-chat-and-file-retention-policies-in-chatgpt) - [Microsoft Security: Guarding AI memory](https://www.microsoft.com/en-us/security/blog/2026/06/22/guarding-ai-memory/) - [Proton: Data Retention Policy](https://proton.me/business/blog/data-retention-policy) ## Questions about AI Memory Retention and Deletion Policy for a Small Business ### If someone asks you to forget specific data, how do you handle that request? Trace the record across raw sources, uploads, summaries, indexes, caches, logs, backups, connectors, and provider-side memory. Delete or expire each eligible copy under the approved policy, record exceptions or legal holds, and use a canary test to prove the data no longer returns. ### What happens to connected app data when I delete a chat? A chat deletion does not by itself prove that connected, derived, indexed, or backed-up copies were removed. The business policy should inventory each dependency, name the owner and deletion trigger, and verify every eligible store separately. ### How do we make an agent forget? Set retention by record type and named business purpose instead of applying one blanket period to all memory. Remove the source and every eligible derived copy, preserve required exceptions, then search with a synthetic canary to test whether retrieval still exposes it. ## Save the visual summary Inventory raw, derived, indexed, and backed-up memory | Keep records for a named reason | Set retention by record type | Separate facts, assumptions, preferences, and action logs [Download the PNG](/infographics/ai-memory-retention-and-deletion-policy-1200x1500.png) ## Related: Owned Memory - [AI Agent Security Risks for a Small Business: Test Access, Actions, and Evidence](/articles/ai-agent-security-risks-small-business) - [Choose local or cloud AI memory by who controls the data](/articles/is-ai-memory-safe) - [What Client Data Belongs in AI Agent Memory?](/articles/client-data-in-ai-agent-memory) ---- ```text .|########||. .|########||. .|########||. |##||. .||##|. |##||. .||##|. |##||. .||##|. |#|. .|#|. |#|. .|#|. |#|. .|#|. |#| |#| |#| |#| |#| |#| .#| |#. .#| |#. .#| |#. .#. .#| .#. .#| .#. .#| |#. .#| |#. .#| |#. .#| |# ...... #| |# ...... #| |# ...... #| .# ||#########| #| .# ||#########| #| .# ||#########| #| .#. |######||######|. .#. .#. |######||######|. .#. .#. |######||######|. .#. #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# # .||####|###########################|. # # .||####|###########################|. # # .||####|###########################|. # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || #. .######|#|#########| .# #. .######|#|#########| .# #. .######|#|#########| .# .#. |#####||||#####| .#. .#. |#####||||#####| .#. .#. |#####||||#####| .#. |# ||########|| #. |# ||########|| #. |# ||########|| #. |# ...... #| |# ...... #| |# ...... #| |#. .#| |#. .#| |#. .#| |#. .#. |#. .#. |#. .#. .#| |#. .#| |#. .#| |#. |#| |#| |#| |#| |#| |#| .|#|. .|#|. .|#|. .|#|. .|#|. .|#|. |##||. .||##| |##||. .||##| |##||. .||##| .||########||. .||########||. .||########||. Scale With Search 2026 [scalewithsearch.com](https://scalewithsearch.com) ```