Assay / Solutions / Pharma
Industry · Pharmaceutical

Agent Trust for Pharmaceutical Manufacturing

In a regulated process the question is never just whether the work was done. It is who did it, what qualified them, and can you show that to an inspector. Assay makes that answer a record rather than a recollection.


The problem

Agents act in validated processes without a qualification record

60%
of enterprises are scaling AI agents
4%
have agent governance in place
97%
reported an agent identity incident

Source: 2025 industry surveys of enterprise AI agent adoption.

Pharmaceutical manufacturing runs on documented qualification. People are trained and assessed, equipment is qualified, methods are validated, and every one of those carries evidence a regulator can inspect.

Agents entering environmental monitoring, deviation triage, batch review support and supplier coordination arrive with no equivalent. There is no qualification record, no competence evidence, and often no durable statement of which agent acted at all.

Data integrity expectations do not soften because the actor is software. Attributable, legible, contemporaneous, original, accurate applies to an agent's actions as much as a technician's — and “attributable” is precisely what an unregistered agent cannot satisfy.


The solution

Attributable identity and evidenced competence

Every agent carries a durable, portable identity. Every outcome is attributable to that identity, timestamped, optionally bound to an evidence hash, and recorded with the counterparty that confirmed it. Attribution stops being a gap.

Assay ships a pharma preset so scoring reflects the risk posture of a regulated process rather than generic defaults, and the model that computed a score travels with it on the passport — so a reviewer can see which rules applied.

Capabilities are evidenced separately, which maps naturally onto qualification thinking: an agent qualified for one task class is not thereby qualified for another, and the record shows the distinction instead of hiding it.

pharma presetattributable actionsevidence hashestimestampedper-capability qualificationsigned passports

Why it holds up

What you actually get

Attributable by construction

Every outcome ties to a durable agent identity with a timestamp and an optional evidence hash. Who acted is a stored fact.

Qualification-shaped scoring

Independent per-capability evidence maps onto how qualification already works: qualified for this task class, not qualified for that one.

Transparent model

Preset parameters are public and inspectable. A scoring model you cannot explain to a reviewer is a model you cannot use in a regulated process.

Evidence hashes

Bind an attestation to a hash of the underlying record so the outcome and the evidence can be tied together later without Assay holding your data.


Live API

Inspect the pharma trust model, verify an agent

The model is public — useful when validation asks how the number is computed.

# Public. Every industry trust model and its tunable parameters. curl -s https://assay.foundrynet.io/v1/admin/trust-presets { "presets": { "manufacturing": { "recency_half_life_days": 30, "failure_penalty": 2.5, // harsher than default 2.0 "counterparty_bonus": 1.5, "verification_ceiling": 0.75, "min_verification_rate": 0.15, "velocity_min_gap_seconds": 300, "velocity_discount": 0.8, "volume_k": 8 }, "pharma": { … }, "financial_services": { … }, "logistics": { … }, "energy": { … }, "default": { … } } } # 2. Verify it before you delegate. This is the allow/deny call. curl -s https://assay.foundrynet.io/v1/verify \ -H 'X-API-Key: YOUR_ASSAY_KEY' \ -H 'Content-Type: application/json' \ -d '{"agent_id":"agt_4411f105…", "claimed_capabilities":["deviation_triage", "environmental_monitoring"], "min_trust_threshold":0.7}' { "verified": true, "trust_score": 0.8, "capabilities_confirmed": ["deviation_triage", "environmental_monitoring"], "attestation_count": 8, "last_attestation_age_seconds": 41 } # Gate on BOTH fields. `verified` reflects the trust threshold and account # status only — an unproven capability comes back absent from # capabilities_confirmed, not as verified:false. if verified and "deviation_triage" in capabilities_confirmed: dispatch()
Read the gate carefully. verified reflects the trust threshold and account status. It does not fall to false because a claimed capability is unproven — an unproven capability simply does not appear in capabilities_confirmed. Gate on both fields.

Questions

Frequently asked

Is Assay a validated GxP system?

Assay is infrastructure, not a validated application, and we will not claim otherwise. It provides attributable identity, evidenced competence and durable outcome records that support your validation approach. Qualification of the overall computerised system remains yours.

Does Assay store our batch or process data?

No. Attestations carry a task type, an outcome, an optional counterparty and an optional evidence hash you compute. The underlying records stay in your systems.

How does this map to Part 11 or Annex 11 thinking?

It contributes to attributability and to competence evidence for automated actors. It is one input to a compliance position, not a compliance product, and should be assessed as such by your quality function.

What does the pharma preset change?

It tunes recency, failure penalty, verification requirements and volume sensitivity for a low-tolerance regulated setting. Exact parameters are public at /v1/admin/trust-presets.


Related

Keep reading