Chain-of-Custody for Samples Governance requires that every AI agent involved in the handling, transport, analysis, or disposition of biological samples maintains a continuous, tamper-evident record of custody from sample acquisition through final disposition, including every transfer of physical or logical possession between human operators, automated systems, and storage facilities. Biological samples — including tissue specimens, microbial isolates, synthesised nucleic acids, cell cultures, and environmental samples — carry risks ranging from personal data exposure (genomic privacy) to biosecurity threats (dual-use pathogen material), making unbroken provenance tracking a prerequisite for safety, regulatory compliance, and scientific reproducibility. Without governed chain-of-custody, an organisation cannot verify that a sample reaching an analysis pipeline is the same sample that was collected, that it has been stored under required conditions throughout its journey, or that custody gaps have not introduced contamination, substitution, or diversion of material with dual-use potential.
Scenario A — Robotic Sample Handler Loses Custody Continuity During Facility Transfer: A pharmaceutical company operates two biosafety-level-2 (BSL-2) laboratories 14 kilometres apart. An AI-driven robotic agent manages sample logistics — selecting samples for inter-facility transfer, generating shipping manifests, and updating the laboratory information management system (LIMS). On 12 March, the agent generates a transfer manifest for 38 patient-derived tissue samples scheduled for genomic sequencing at the second facility. The robotic handler loads the samples into a climate-controlled transport container. During transit, the transport vehicle's IoT telemetry link drops for 47 minutes due to a cellular coverage gap. The agent's custody record shows a 47-minute void with no temperature, location, or access-event data. Upon arrival, the receiving laboratory scans the container and records 38 samples — but the custody gap means there is no verifiable evidence that the container was not opened, that samples were not substituted, or that temperature excursion did not occur during the void. The laboratory proceeds with sequencing regardless, treating the gap as a telemetry glitch. Four weeks later, sequencing results for 3 of the 38 samples show unexpected genomic variants inconsistent with the patients' clinical profiles. An investigation reveals that 3 samples were cross-contaminated by a container breach during transit — a breach that occurred during the 47-minute custody void.
What went wrong: The agent had no protocol to flag custody gaps as integrity-threatening events. The 47-minute telemetry void was treated as a connectivity issue rather than a chain-of-custody break. No hold-for-investigation procedure existed for samples arriving with incomplete custody records. The receiving laboratory's acceptance of samples without complete custody documentation meant the contamination was discovered only after sequencing resources worth £92,000 were consumed and 3 patients received erroneous genomic results, requiring clinical follow-up costing £41,000 and triggering a reportable adverse event under the UK Human Tissue Act. Consequence: £133,000 in direct costs, 3 patients recalled for re-testing, regulatory notification to the Human Tissue Authority, and a 6-week suspension of inter-facility sample transfers pending remediation.
Scenario B — AI Agent Authorises Release of Select Agent Without Dual-Custody Verification: A government biosecurity research facility maintains a collection of Tier 1 Select Agents regulated under the US Federal Select Agent Program (42 CFR Part 73). An AI agent manages inventory and access requests for the collection. A researcher submits a request for a vial of a reconstructed 1918 influenza virus for an approved research protocol. The agent validates the researcher's credentials, confirms the research protocol approval, and generates a release authorisation. However, the agent's workflow omits the dual-custody verification required by the facility's Select Agent security plan — two authorised individuals must independently verify and witness the physical transfer. The agent logs a single-person release event and updates inventory accordingly. The error is discovered 6 days later during a routine CDC/APHIS inspection when the inspector notes that the release record shows only one custodian signature rather than the required two. The inspector issues a finding of non-compliance. A full inventory reconciliation is ordered, requiring 72 hours of facility lockdown and verification of all 1,247 vials in the collection. Total cost of the reconciliation and remediation is $890,000, and the facility's Select Agent registration is placed on probationary status for 12 months.
What went wrong: The agent's release workflow was configured with single-custodian authorisation, violating the dual-custody requirement mandated by the Select Agent security plan and 42 CFR Part 73. No validation check existed to confirm that the number of independent custody witnesses matched the security plan requirement. The agent treated credential verification as sufficient for release, without enforcing the physical co-presence requirement. Consequence: $890,000 in reconciliation costs, 72-hour facility lockdown halting all research operations, 12-month probationary registration status, and reputational damage with the CDC Select Agent Program.
Scenario C — Cross-Border Tissue Sample Shipment Loses Provenance at Customs Handover: A clinical genomics company in Germany ships 120 tumour biopsy samples to a sequencing partner laboratory in Singapore for whole-genome sequencing under a clinical trial protocol. An AI agent manages the logistics workflow, generating export documentation, customs declarations, and chain-of-custody records. At Frankfurt customs, the shipment is held for 9 hours for inspection. During the hold, customs officers open the outer packaging to verify contents against the declaration. The customs inspection is documented in the customs system but not in the agent's chain-of-custody record — the agent has no integration with customs inspection systems and records only the departure scan at the originating facility and the arrival scan at the destination. The 9-hour gap in the agent's custody record, combined with evidence of package opening, means the integrity of the cold chain and the identity of the samples cannot be verified for the period of customs inspection. The sequencing partner accepts the shipment and processes 118 of 120 samples (2 vials show visible thawing damage). Results from the 118 samples are submitted to the clinical trial database. Three months later, a regulatory audit by Germany's Federal Institute for Drugs and Medical Devices (BfArM) identifies the custody gap and determines that the 118 sequencing results cannot be attributed to verified samples with unbroken provenance. The BfArM orders 118 results excluded from the clinical trial analysis. The trial sponsor must recall and re-biopsy 118 patients — a process costing approximately €1.4 million and delaying the trial by 8 months.
What went wrong: The agent had no mechanism to incorporate third-party custody events (customs inspections) into its chain-of-custody record. The cross-border handover created a provenance gap that the agent could not detect or flag. No protocol required verification of custody continuity at the destination before sample processing began. The absence of custody documentation for the customs hold period was not treated as a disqualifying event for clinical-grade samples. Consequence: €1.4 million in re-biopsy and re-sequencing costs, 8-month clinical trial delay, 118 patients subjected to repeat invasive procedures, and regulatory finding for inadequate sample traceability under Good Clinical Practice (GCP).
Scope: This dimension applies to any AI agent that initiates, authorises, manages, tracks, or participates in the handling of biological samples — from initial collection or acquisition through storage, transfer, analysis, and final disposition (including destruction or archival). The scope encompasses physical samples (tissue, blood, microbial isolates, synthesised oligonucleotides, cell lines, environmental samples) and their digital representations (associated metadata, sequence data, and analytical results that must be traceable to a specific physical sample). The dimension applies regardless of whether the agent directly manipulates samples (robotic agents) or manages sample workflows from a supervisory position (logistics agents, LIMS agents). It covers intra-facility transfers, inter-facility transfers, cross-border shipments, and transfers between organisations. Organisations operating under the Select Agent Program, the Nagoya Protocol, human tissue legislation, Good Clinical Practice, or Good Manufacturing Practice face heightened regulatory obligations that this dimension explicitly addresses. The scope excludes purely computational operations on sequence data that have already been de-linked from physical samples — such operations are governed by AG-709 (Sequence Data Sensitivity Governance).
4.1. A conforming system MUST maintain a continuous, tamper-evident chain-of-custody record for every biological sample under its management, recording at minimum: sample unique identifier, custodian identity at each transfer point, timestamp of each custody event (acquisition, transfer, storage, retrieval, analysis, disposition), environmental conditions during custody (temperature, humidity where applicable), and the physical location of the sample at each recorded event.
4.2. A conforming system MUST record every custody transfer event — including transfers between human operators, between human operators and automated systems, between automated systems, and between facilities — with both the releasing and receiving custodian identities cryptographically bound to the transfer record within 60 seconds of the physical transfer.
4.3. A conforming system MUST detect and flag any custody gap — a period exceeding a configurable threshold (default: 15 minutes) where no custody event has been recorded for a sample that is not in verified static storage — and generate an automated alert to the designated biosafety or quality officer within 30 minutes of gap detection.
4.4. A conforming system MUST enforce sample acceptance controls at every receiving point (facility, laboratory, analytical instrument), rejecting or quarantining any sample whose chain-of-custody record is incomplete, contains unresolved gaps, or shows evidence of tampering, and escalating to a human authority for disposition decisions per AG-019.
4.5. A conforming system MUST enforce custody requirements specific to the sample's biosafety and regulatory classification, including dual-custody verification for Select Agents and Toxins (or jurisdictional equivalent), witnessed destruction for biohazardous waste, and documented consent verification for human-derived samples, and MUST reject any custody workflow that does not satisfy the classification-specific requirements.
4.6. A conforming system MUST integrate environmental monitoring data (temperature, humidity, light exposure, vibration) into the chain-of-custody record so that any environmental excursion during custody is linked to the specific samples affected and triggers a quality hold preventing further processing until a qualified human reviews and dispositions the affected samples.
4.7. A conforming system MUST retain chain-of-custody records for the full retention period required by the applicable regulatory framework — a minimum of 10 years for clinical samples, 7 years for research samples under Select Agent regulations, and 5 years for all other samples — with records stored in an immutable, append-only format consistent with AG-055.
4.8. A conforming system SHOULD implement cryptographic sealing of custody records using hash chains or equivalent mechanisms, such that any retroactive alteration of a custody record is computationally detectable and any record can be independently verified against the chain.
4.9. A conforming system SHOULD integrate with third-party custody systems (customs, courier services, external laboratories) via standardised data exchange protocols to eliminate provenance gaps at organisational boundaries, or SHOULD implement a reconciliation protocol at each boundary crossing that verifies custody continuity before sample acceptance.
4.10. A conforming system SHOULD implement sample identity verification at each custody transfer point using physical verification mechanisms (barcode scanning, RFID, weight verification, visual confirmation) to detect sample substitution, mislabelling, or loss.
4.11. A conforming system MAY implement blockchain-based or distributed-ledger custody records for multi-organisation sample workflows where no single organisation has authoritative control over the complete custody chain.
4.12. A conforming system MAY implement predictive custody risk scoring that assesses the likelihood of custody breaks based on route characteristics, transit duration, environmental forecasts, and historical incident data, and routes samples through lower-risk pathways when alternative options exist.
Biological samples are simultaneously scientific assets, regulated materials, and — in the case of human-derived specimens — extensions of individual privacy and autonomy. The chain-of-custody is the evidentiary foundation that connects an analytical result to the specific biological material from which it was derived, and connects that material to the individual, organism, or environment from which it was collected. When the chain-of-custody breaks, the link between result and source is severed, and the scientific, clinical, and legal value of the result collapses.
The threat model for chain-of-custody failures in AI-managed sample workflows encompasses five distinct categories. First, accidental provenance loss: robotic handlers, automated storage systems, and logistics agents operate at speeds and scales that exceed human capacity for manual record-keeping. A robotic arm that moves 200 samples per hour generates 200 custody events per hour — any event that is not recorded in real time is lost. Unlike human operators who can reconstruct events from memory (albeit unreliably), AI agents have no retrospective recall of unrecorded events. If the recording mechanism fails, the custody record has a gap that cannot be filled after the fact. Second, environmental excursion concealment: samples that require cold-chain maintenance (the majority of biological samples) are vulnerable to temperature excursions during transport and storage transitions. Without environmental data integrated into the custody record, an excursion may degrade a sample without any visible indicator, and the resulting analytical data will be unreliable without any flag in the record. Third, sample substitution or diversion: in settings where samples have economic value (clinical trial samples), regulatory significance (Select Agents), or dual-use potential (pathogen isolates), the chain-of-custody serves as a security control against intentional substitution or diversion. A custody gap is an opportunity for substitution. Fourth, cross-boundary provenance gaps: when samples cross organisational boundaries — between a clinic and a laboratory, between two national jurisdictions, through a customs checkpoint — the custody record typically transitions between systems. Each system transition is a gap risk, because the releasing system records the release and the receiving system records the receipt, but the interval between is often unrecorded. AI agents that manage only one side of a boundary crossing cannot provide end-to-end custody assurance without explicit integration or reconciliation protocols. Fifth, regulatory non-compliance escalation: biological sample handling is among the most heavily regulated domains in science and medicine. The Select Agent Program, the Nagoya Protocol on Access and Benefit-Sharing, the EU Tissue and Cells Directive (2004/23/EC), Good Clinical Practice (ICH E6), and Good Manufacturing Practice all mandate traceability of biological materials. A chain-of-custody failure is not merely a quality issue — it is a regulatory violation that can result in facility closure, criminal prosecution (for Select Agent violations), and exclusion of clinical data from regulatory submissions.
AI agents introduce both new capabilities and new risks to chain-of-custody management. The capability is continuous, high-resolution tracking that exceeds human capacity — an AI agent can record every custody event in real time with environmental telemetry, creating richer provenance records than any manual system. The risk is that the agent's recording is only as reliable as its sensors, its connectivity, and its programming. A human operator who moves a sample knows they moved it, even if they fail to record the movement. An AI agent that loses connectivity does not know what happened during the connectivity gap — and cannot reconstruct it. The governance of chain-of-custody must therefore address the reliability of the recording mechanism itself, not just the content of the records.
Chain-of-custody governance for AI-managed biological samples requires an architecture that captures custody events in real time, integrates environmental telemetry, detects and flags gaps, and enforces acceptance controls at every handover point. The architecture must be resilient to connectivity failures, sensor malfunctions, and cross-system boundary transitions, because these are the conditions under which custody breaks most commonly occur.
Recommended patterns:
Anti-patterns to avoid:
Clinical Trials and Good Clinical Practice. ICH E6(R2) Good Clinical Practice requires traceability of all biological samples collected under a clinical trial protocol. Chain-of-custody gaps can disqualify sample-derived data from regulatory submissions, potentially invalidating clinical trial results. Organisations conducting GCP-regulated trials should implement the full custody architecture with environmental telemetry and boundary reconciliation, and should ensure that custody records meet the requirements of 21 CFR Part 11 (electronic records) or EU Annex 11 (computerised systems) as applicable.
Select Agent Facilities. Facilities registered under the US Federal Select Agent Program face the most stringent custody requirements. Dual-custody (two-person rule) for Tier 1 agents is not discretionary — it is a condition of registration. AI agents operating in Select Agent facilities must enforce dual-custody at the workflow level, not merely recommend it. Violations can result in facility closure, criminal referral to the FBI, and revocation of Select Agent registration.
Cross-Border Biobanking. Organisations that ship samples across national borders must contend with the Nagoya Protocol (access and benefit-sharing for genetic resources), CITES (for samples from protected species), and national customs requirements. Each border crossing introduces a custody gap risk. Cross-border biobanking networks should implement federated custody ledgers that allow each participating institution to contribute events to a shared provenance record with cryptographic verification.
Pharmaceutical Manufacturing. Good Manufacturing Practice (GMP) requires that raw materials — including biological starting materials — are traceable from receipt through use in manufacturing. AI agents managing pharmaceutical supply chains must maintain custody records that satisfy GMP Annex 11 requirements and support recall operations that trace a finished product batch back to its constituent biological materials.
Basic Implementation — The organisation maintains chain-of-custody records for all biological samples, recording custodian identity, timestamps, and sample identifiers at each transfer point. Records are stored in a tamper-resistant format. Custody gaps trigger manual investigation. Classification-specific requirements (dual-custody for Select Agents, consent verification for human samples) are enforced through documented procedures. All mandatory requirements (4.1 through 4.7) are satisfied through a combination of automated and manual controls.
Intermediate Implementation — All basic capabilities plus: the custody ledger is cryptographically sealed with hash chains enabling tamper detection. Environmental telemetry is integrated into custody records with automated excursion detection and quality holds. Gap detection is fully automated with configurable thresholds per sample type. Dual-confirmation transfers are implemented for all custody handovers. Boundary reconciliation protocols operate at every organisational interface. Sample identity verification (barcode, RFID, or weight) is performed at every transfer point. Custody metrics (gap frequency, gap duration, exception rate) are reported to quality management quarterly.
Advanced Implementation — All intermediate capabilities plus: federated custody ledgers support multi-organisation sample workflows with cryptographic verification across institutional boundaries. Third-party custody systems (customs, couriers, external laboratories) are integrated via standardised data exchange protocols, eliminating boundary gaps. Predictive custody risk scoring routes samples through lower-risk pathways. Independent audit annually validates custody record integrity, gap detection sensitivity, and classification-specific enforcement. The custody system is integrated with AG-713 (Biohazard Incident Routing) for automatic incident escalation when custody failures involve hazardous materials.
Required artefacts:
Retention requirements:
Access requirements:
Test 8.1: Custody Record Completeness and Continuity
Test 8.2: Custody Transfer Dual-Confirmation
Test 8.3: Gap Detection and Alerting
Test 8.4: Sample Acceptance Control Enforcement
Test 8.5: Classification-Specific Custody Enforcement
Test 8.6: Environmental Excursion Detection and Quality Hold
Test 8.7: Custody Record Retention and Immutability
| Regulation | Provision | Relationship Type |
|---|---|---|
| US Select Agent Regulations | 42 CFR Part 73 (Select Agents and Toxins) | Direct requirement |
| EU Tissue and Cells Directive | Directive 2004/23/EC, Article 8 (Traceability) | Direct requirement |
| ICH E6(R2) | Good Clinical Practice, Section 8 (Essential Documents) | Direct requirement |
| Nagoya Protocol | Articles 6, 7 (Access and Benefit-Sharing) | Supports compliance |
| EU AI Act | Article 9 (Risk Management System) | Supports compliance |
| 21 CFR Part 11 | Electronic Records, Electronic Signatures | Supports compliance |
| ISO 20387 | Biobanking — General Requirements | Direct requirement |
| UK Human Tissue Act 2004 | Sections 32-34 (Restrictions on Use) | Supports compliance |
The Federal Select Agent Program, jointly administered by the CDC and APHIS, imposes the most stringent chain-of-custody requirements in the biological sciences. Section 73.17 requires registered entities to maintain accurate, current inventory of all Select Agents and Toxins, including records of all access, transfers, and destruction. Section 73.11 requires a security plan that addresses access controls and personnel suitability. For Tier 1 Select Agents — those with the greatest potential for deliberate misuse — the regulations require a two-person access policy (dual custody). AI agents operating within Select Agent facilities must enforce these requirements at the system level. A chain-of-custody failure for a Select Agent is not a quality issue — it is a federal security violation that can result in criminal prosecution under 18 U.S.C. § 175 (Biological Weapons Anti-Terrorism Act), facility closure, and referral to the FBI for investigation. The stakes require that custody controls are not merely recommended but are technically enforced and independently verified.
Article 8 of the Directive requires Member States to ensure that all tissues and cells procured, processed, and distributed on their territory can be traced from the donor to the recipient and vice versa. The Directive mandates a unique identification code (the Single European Code under Commission Directive 2015/565) and traceability records maintained for a minimum of 30 years. AI agents managing tissue and cell workflows within the EU must implement custody records that comply with the Single European Code requirements and support bidirectional traceability. The 30-year retention requirement exceeds the general retention requirements of this dimension and takes precedence where applicable.
Section 8 of GCP requires that essential documents — including records demonstrating the traceability of clinical samples — are maintained throughout the trial and retained for the required period. Section 5.18.4 requires that the sponsor ensure appropriate handling and storage of clinical supplies, including biological samples. A chain-of-custody break for a clinical trial sample can disqualify the sample-derived data from the regulatory submission, because the regulator cannot verify that the analysed material corresponds to the material collected from the enrolled participant. For Phase III trials where each sample represents months of patient participation, the cost of a custody failure is measured not only in direct remediation expenses but in trial delay and potential loss of the entire data point.
The Nagoya Protocol on Access and Benefit-Sharing requires that genetic resources are accessed with prior informed consent and that benefits arising from their utilisation are shared fairly and equitably with the providing country. Chain-of-custody documentation is the evidentiary mechanism that demonstrates compliance with access requirements — proving that a genetic resource was obtained through legitimate channels and can be traced back to its country of origin. AI agents managing genetic resource collections must maintain custody records that support Nagoya Protocol due diligence obligations, as implemented through national legislation such as the EU ABS Regulation (Regulation 511/2014).
For organisations subject to US FDA oversight, electronic chain-of-custody records must comply with 21 CFR Part 11, which requires that electronic records are trustworthy, reliable, and equivalent to paper records. This includes requirements for audit trails that record the date, time, and identity of any person who creates, modifies, or deletes an electronic record. The append-only, cryptographically sealed custody ledger required by this dimension inherently satisfies the Part 11 audit trail requirement, provided that the system also implements the access controls, authority checks, and electronic signature requirements specified in Part 11.
ISO 20387 specifies general requirements for biobanking, including requirements for sample traceability, quality management, and data management. The standard requires that biobanks maintain records sufficient to trace each sample from collection through storage, use, and disposition. AI agents operating within ISO 20387-accredited biobanks must produce custody records that satisfy the standard's traceability requirements and support the biobank's quality management system.
| Field | Value |
|---|---|
| Severity Rating | Critical |
| Blast Radius | Multi-domain — affects scientific validity, patient safety, biosecurity, regulatory compliance, and legal liability across all operations involving biological samples |
Consequence chain: A chain-of-custody failure for biological samples triggers a cascading consequence chain whose severity depends on the sample type and regulatory context. The immediate failure mode is a custody gap or integrity violation — a period or event where sample provenance cannot be verified. The first-order consequence is that the integrity of any analytical result derived from the affected sample is unverifiable — the organisation cannot prove that the result corresponds to the material that was collected. For clinical samples, this means patient results may be attributed to the wrong individual or may reflect degraded material, creating a direct patient safety risk. For Select Agent samples, a custody gap is a security incident that must be reported to the CDC/APHIS and may trigger a federal investigation. The second-order consequence is regulatory: clinical trial data derived from samples with broken custody may be excluded from regulatory submissions, potentially invalidating trial results; Select Agent custody violations can result in facility closure and criminal prosecution; human tissue traceability failures trigger regulatory notifications under the Human Tissue Act or Tissue and Cells Directive. The third-order consequence is operational: sample re-collection (requiring patient recall for clinical samples), re-analysis, facility lockdowns pending inventory reconciliation, and remediation of the custody system. In clinical trial settings, a single custody failure affecting a cohort of samples can delay a trial by 6-12 months at a cost of £500,000 to £5 million depending on trial phase and sample count. For Select Agent facilities, a dual-custody violation can result in permanent revocation of registration, effectively shutting down an entire research programme. The reputational consequence is severe in all cases: collaborators, patients, and regulators lose confidence in the organisation's ability to handle biological materials responsibly, affecting future research partnerships, clinical trial participation, and regulatory goodwill.
Cross-references: AG-001 (Operational Boundary Enforcement) defines the operational boundaries within which the agent manages samples — custody governance ensures that sample handling stays within those boundaries. AG-007 (Governance Configuration Control) governs the configuration of custody workflow rules, classification mappings, and threshold settings. AG-008 (Governance Continuity Under Failure) ensures that custody recording continues during system failures — connectivity loss must not silently create custody gaps. AG-019 (Human Escalation & Override Triggers) defines the escalation pathways invoked when custody gaps, acceptance control failures, or classification violations are detected. AG-029 (Data Classification Enforcement) governs the classification of sample-associated data, which determines custody requirements based on sample sensitivity. AG-042 (Encryption & Cryptographic Control Governance) governs the cryptographic mechanisms used for custody record sealing and tamper detection. AG-043 (Access Control & Credential Governance) ensures that only authorised individuals can serve as custodians and that custodian identity is reliably established at each transfer point. AG-055 (Audit Trail Immutability & Completeness) provides the foundational requirement for immutable, complete audit records that the custody ledger implements. AG-068 (Intellectual Property Boundary Governance) addresses IP boundaries for sample-derived data and ensures that custody transfers between organisations respect IP ownership agreements. AG-210 (Multi-Jurisdictional Regulatory Mapping) provides the jurisdictional mapping necessary to determine which custody requirements apply when samples cross borders — Select Agent regulations in the US, the Tissue and Cells Directive in the EU, the Human Tissue Act in the UK, and the Nagoya Protocol internationally. AG-712 (Material Access Governance) controls who may access biological materials — custody governance tracks what happens after access is granted. AG-713 (Biohazard Incident Routing Governance) defines incident routing when custody failures involve biohazardous materials.