At Gekko Photonics, we design and manufacture process Raman analyzers in Poland — in inline, laboratory, and portable variants — and we observe the regulatory wave of Q1 2026 from the perspective of teams currently installing PAT in both pharmaceuticals and specialty chemicals. The first three months of this year brought an agreement on the EU Pharma Package, joint FDA and EMA guidelines for AI/ML in drug development, the finalization of consultations on Annex 22, and a year of full compliance with the new Annex 1. Below is a condensed overview of the changes that QbD and PAT teams should have on their desks.

EU Pharma Package — political agreement and final texts

On December 11, 2025, the European Parliament and the EU Council reached a political agreement on the reform of EU pharmaceutical law — the most profound in over two decades. On March 6, 2026, the Council published the final compromise texts of the package (a new directive on the Community code relating to medicinal products for human use and a regulation on authorization and supervision procedures). Formal adoption by both institutions is expected in summer 2026, after which a transition period until 2028 will begin, during which member states will adapt national law.

The package itself does not directly change the PAT/QbD framework — it is not a document on process analysis. However, it reconfigures the landscape in which PAT operates: new authorization frameworks, regulatory data protection, sustainability, and mitigating drug shortages. Real-time release testing and continuous manufacturing fit into the narrative of „accelerating access to medicines while maintaining quality” that the reform promotes.

Joint FDA and EMA guidelines for AI/ML in drug development (January 2026)

In January 2026, the FDA and EMA published 10 common principles on the use of artificial intelligence and machine learning in the development of medicinal products. The principles emphasize validation, transparency, bias control, model lifecycle management, and a risk-based approach to documentation.

From the perspective of PAT implementation teams, this is an important signal: multivariate chemometric models (PLS, PCA, convolutional networks) fall under the same logic as broader AI models — requiring documented definition of intended use, training data set, acceptance criteria, and post-deployment drift monitoring. If someone is implementing Raman with PLS calibration as CQA monitoring in GMP, these principles define the direction in which inspections are heading.

Annex 22 — AI in GMP in the finalization phase

The draft EU GMP Annex 22 (Artificial Intelligence) was released for public consultation on July 7, 2025, and the comment period closed on October 7, 2025. Finalization, along with the related revision of Annex 11 (Computerized Systems), is expected in mid-2026.

Importantly — Annex 22 is limited to static, deterministic AI models in GMP-critical decisions; adaptive and generative models are explicitly excluded from such applications. Every model within the scope of Annex 22 must have documented intended use, training data, validation, acceptance criteria, and a maintenance plan.

Practical consequence for Raman chemometrics: typical PLS and PCA models, calibrated on a closed data set and frozen after validation, fall into the „static, deterministic” category. Models based on convolutional neural networks (CNNs), while permissible, require more careful documentation of the training set, validation on an independent test set, and justification for why the model remains deterministic in production after training.

ICH Q14 + Q2(R2) — implementation matures

ICH Q14 (Analytical Procedure Development) came into effect in the EU on June 14, 2024, and in 2026 it enters a period of first mature implementations in pharmaceutical organizations. Key contribution of Q14: AQbD (Analytical Quality by Design) with three phases of the method lifecycle — Procedure Design, Procedure Performance Qualification, and Continued Procedure Performance Verification.

Complementary ICH Q2(R2) — updated validation of analytical procedures — explicitly extends its scope to spectroscopic techniques with chemometric models: Raman, NIR, NMR. This fills a gap that for years generated regulatory questions for submissions involving multivariate methods. Raman validation under the Q2(R2) paradigm proceeds in two phases: building a calibration model on a training set and verifying predictions on a test set.

For teams that learned Raman validation „under” the old Q2 — this is not cosmetic, but a redesign of the validation report. New sections include, among others, model range, robustness against raw material variance, drift monitoring, and a re-calibration procedure. Q14 adds a lifecycle requirement: a method that works one year after implementation must continue to be monitored for performance.

Annex 1 — 2026 as the year of full compliance

The new Annex 1 of EU GMP (Manufacture of Sterile Medicinal Products) came into effect on August 25, 2023. The industry had over two years for full adaptation, and 2026 is widely regarded as the year of full compliance. Inspectors from the EMA and national agencies in 2026 are scrutinizing the implementation of critical requirements: a comprehensive Contamination Control Strategy (CCS), continuous environmental monitoring in Grade A, and integrated Quality Risk Management at every stage of the sterile process.

For PAT, this is a significant context — Annex 1 does not explicitly require Raman, but it develops the philosophy of a „continuous state of control,” in which process analytics (real-time monitoring of product quality, not just air particles) becomes a natural element of the CCS. Sterile drug manufacturers are increasingly seeking spectroscopic tools for continuous verification of raw materials, processing media, and the final product.

What this means beyond pharmaceuticals — transferring the PAT regime to chemistry

Specialty chemistry, polymer chemistry and formulation chemistry have long observed pharmaceutical PAT/QbD as a benchmark. Three areas where 2026 strengthens this transfer:

• REACH requirements and subsequent amendments to the CMRD increase pressure for documented monitoring of carcinogenic, mutagenic, and reprotoxic substances in chemical processes — in-line analytics is increasingly expected, not just point laboratory measurements.
• Lifecycle of chemometric models — chemical plants that adopt PAT solutions from pharmaceuticals also copy pharmaceutical documentation patterns: validation on an independent test set, drift monitoring, traceability of every model decision.
• PAT-by-design for new installations — in European investment projects in specialty chemicals (resins, composites, surfactants), we increasingly encounter schedules where process analytics is part of the P&ID at the FEED stage, rather than an addition at the end.

In other words — what is a regulatory obligation for pharmaceuticals becomes a competitive tool and a path to compliance with EU environmental and occupational safety requirements for chemistry.

Gekko Photonics solutions for PAT and QbD teams

At Gekko Photonics, we deliver a full PAT stack based on Raman spectroscopy — from a probe in the reactor to an analytical platform. We configure it for specific compliance requirements, regardless of industry.

• Spectrally X1 INLINE — process Raman analyzer with an immersion probe, continuous measurement in a reactor or pipeline, PROFIBUS/PROFINET communication, optional Retractex self-cleaning probe for difficult media. It fits into the continuous state of control paradigm from Annex 1 and ICH Q13.
• Spectrally X1 LAB — benchtop analyzer with a carousel for up to 25 samples and through-package analysis via transparent packaging, ideal for validating chemometric models before inline implementation (Procedure Performance Qualification phase in Q14 terminology).
• Spectrally X1 PORTABLE — portable analyzer for incoming QC, raw material identification, and mobile model verification in the field, IP54, ready for use outside the laboratory.
• Spectrally OS — software layer common to the entire X1 family: PLS, PCA, and CNN models, a library of approximately 28,000 reference spectra, CSV/PDF/RAW export, role-based access control, reporting, and model drift monitoring — i.e., the tools required by Continued Procedure Performance Verification from Q14 and lifecycle documentation from Annex 22.

The complete offering of our process analyzers includes configuration for a specific application, including probes, fiber optic lengths, chemometric models, and integration with the client's DCS/MES. We also perform a feasibility study on the client's samples ourselves — this is our default step zero, before the client approves CAPEX.

FAQ

How does PAT differ from traditional quality control?

Traditional quality control verifies the product after the fact — by taking samples from the line or batch and performing measurements in the laboratory. PAT shifts the measurement moment into the process in real time, measuring critical process parameters (CPP) and critical quality attributes (CQA) on-line or at-line, so that decisions about the process course can be made during manufacturing, not after its completion. This is the foundation of the QbD philosophy and ICH Q13, but for several years it has also become a standard in specialty chemicals.

How do ICH Q14 and Q2(R2) affect Raman method validation?

ICH Q2(R2) explicitly covers the validation of spectroscopic methods with chemometric models, including Raman. In practice, this means a two-stage validation: building a calibration model on a training set and independent verification of predictions on a test set. ICH Q14 adds AQbD and a method lifecycle requirement — documented intended use, working range, acceptance criteria, post-implementation performance monitoring, and a re-calibration procedure.

Does Annex 22 apply to all chemometric models?

Annex 22 covers AI models used in GMP-critical decisions, limiting itself to static, deterministic models. Standard PLS and PCA models, calibrated on a closed set and frozen after validation, fall into this category. Adaptive models (learning continuously from new data) and generative models are explicitly excluded from GMP-critical decisions. Models based on convolutional networks are permissible but require more careful validation documentation and justification that they remain deterministic after training.

Does Gekko Photonics provide PAT solutions for pharmaceuticals?

Yes. At Gekko Photonics, we configure the Spectrally X1 family (INLINE, LAB, PORTABLE) along with the Spectrally OS platform to meet the requirements of pharmaceutical PAT and QbD teams. Chemometric models (PLS, PCA, CNN), a library of approximately 28,000 reference spectra, method lifecycle documentation, RBAC, and auditable report export — these are elements that directly support validation under the ICH Q2(R2) and Q14 paradigm and the documentation required by Annex 22. We assist clients both in the feasibility phase and during inline implementation in the reactor.

When is full compliance with the new Annex 1 required?

The new Annex 1 (Manufacture of Sterile Medicinal Products) came into effect on August 25, 2023, and 2026 is widely regarded as the year of full compliance. Inspectors in 2026 will closely examine the implementation of the Contamination Control Strategy, continuous monitoring in Grade A, and the integration of Quality Risk Management in sterile processes.

What's next — a short action map for Q2 2026

For pharmaceutical teams: finalizing full compliance with Annex 1, reviewing validation documentation for analytical methods under ICH Q14/Q2(R2), and preparing an inventory of AI/ML models for the upcoming Annex 22. For specialty chemicals teams: assessing whether current measurement methods will withstand upcoming environmental and occupational safety tightening, and considering PAT-by-design in new investment projects.

At Gekko Photonics, we tailor the Spectrally X1 configuration to the specific process and regulatory requirements — from selecting the probe wavelength, through chemometric models, to integration with the DCS/MES stack. We invite you to a 30-minute conversation with an application engineer, in which we will discuss your specific application. In a standard scenario, we perform a test measurement on the client's samples in our laboratory within 2 weeks of signing an NDA, and a full inline implementation typically takes from 3 to 5.5 months from the kick-off workshop to an operational system in the process. If the regulatory topic is just emerging, we also recommend our decision-maker's guide to process Raman.