The phar­ma­ceu­ti­cal in­dus­try is un­der­go­ing a sig­nif­i­cant trans­for­ma­tion, dri­ven by the con­ver­gence of dig­i­tal tech­nolo­gies and ad­vanced man­u­fac­tur­ing. Among these, dig­i­tal twins stand out as a piv­otal en­abler of the “CPV of the Fu­ture” ini­tia­tive, an in­ter­na­tion­al, cross-dis­ci­pli­nary project led by Aizon co-founder and Chief Sci­ence Of­fi­cer Toni Man­zano, on be­half of the PDA and oth­er lead­ing in­sti­tu­tions. This trail­blaz­ing project pi­o­neered the cre­ation and use of a dig­i­tal twin to man­age a bio­phar­ma­ceu­ti­cal man­u­fac­tur­ing process work­ing un­der GxP con­di­tions. With this mod­el, Con­tin­u­ous Process Ver­i­fi­ca­tion (CPV) evolves from a com­pli­ance oblig­a­tion in­to a proac­tive, da­ta-dri­ven prac­tice.

At Aizon, we em­pow­er these in­no­v­a­tive ap­pli­ca­tions with our GxP-com­pli­ant plat­form, which brings dig­i­tal twin tech­nol­o­gy to life in reg­u­lat­ed en­vi­ron­ments, help­ing man­u­fac­tur­ers op­ti­mize qual­i­ty, re­duce de­vi­a­tions, and stay ahead of com­pli­ance de­mands.

Not all dig­i­tal twins are cre­at­ed equal, and, in fact, in­dus­try lead­ers ad­vise against lump­ing what are very dif­fer­ent sys­tems in­to the same cat­e­go­ry. At the re­cent 2025 PDA Good Dig­i­tal Man­u­fac­tur­ing Con­fer­ence in Basel, a broad con­sen­sus re­in­forced a tax­on­o­my first pro­posed by Kritzinger et al. in 2018 that clar­i­fies the dif­fer­ences be­tween three dig­i­tal con­structs:

• Dig­i­tal Mod­els: Sta­t­ic sim­u­la­tions with no live da­ta con­nec­tion.
• Dig­i­tal Shad­ows: Re­al-time da­ta feeds with­out feed­back loops.
• Dig­i­tal Twins: Re­al-time, bidi­rec­tion­al sys­tems that can sim­u­late, pre­dict, and in­flu­ence phys­i­cal process­es.

This dis­tinc­tion em­pha­sizes that the de­nom­i­na­tion of ‘dig­i­tal twin’ should be re­served for sys­tems that are ca­pa­ble of au­tonomous­ly mak­ing ad­just­ments, there­fore clos­ing the loop be­tween the dig­i­tal and phys­i­cal realms. Full-fledged dig­i­tal twins are more than sim­u­la­tions. They are dy­nam­ic, bidi­rec­tion­al re­al-time dig­i­tal rep­re­sen­ta­tions of phys­i­cal sys­tems with the ca­pac­i­ty to act au­tonomous­ly.

Fur­ther­more, each of these con­structs can be built with or with­out ar­ti­fi­cial in­tel­li­gence, and may or may not be GMP-com­pli­ant, de­pend­ing on how they are de­signed and de­ployed. Aizon en­ables the cre­ation of true GMP dig­i­tal twins with em­bed­ded com­pli­ance fea­tures such as trace­abil­i­ty, mod­el val­i­da­tion, and da­ta in­tegri­ty con­trols, with the pos­si­bil­i­ty of be­ing pow­ered by AI to ob­tain re­al-time pre­dic­tive in­sights.

Many cur­rent phar­ma­ceu­ti­cal ap­pli­ca­tions, such as re­al-time mon­i­tor­ing of CPPs and CQAs, pre­dic­tive main­te­nance, and au­dit-ready da­ta col­lec­tion, are well served by dig­i­tal shad­ows, sys­tems that re­flect process da­ta in re­al time but don’t in­ter­vene. These sys­tems pro­vide valu­able vis­i­bil­i­ty, help­ing man­u­fac­tur­ers pre­vent de­vi­a­tions, re­duce down­time, and stream­line com­pli­ance.

How­ev­er, as we have seen, a true dig­i­tal twin goes fur­ther. With bidi­rec­tion­al da­ta ex­change and au­tonomous de­ci­sion-mak­ing, dig­i­tal twins close the loop be­tween de­tec­tion and ac­tion.

This means that a dig­i­tal twin can not on­ly iden­ti­fy a drift in flow rate, but au­to­mat­i­cal­ly ad­just it with­in val­i­dat­ed ranges. It doesn’t just flag po­ten­tial yield loss, but trig­gers cor­rec­tive changes in re­al time.

This au­ton­o­my en­ables faster, smarter, and more con­sis­tent process op­ti­miza­tion, es­pe­cial­ly in com­plex or high-vari­abil­i­ty man­u­fac­tur­ing like bi­o­log­ics or per­son­al­ized ther­a­pies.

Fur­ther­more, a true dig­i­tal twin does­n't ex­ist in iso­la­tion, it can evolve by learn­ing from sim­i­lar twins across the net­work. This di­a­logue be­tween dig­i­tal twins en­ables a form of col­lec­tive in­tel­li­gence: a biore­ac­tor twin at one site, for ex­am­ple, can lever­age in­sights from twins at oth­er sites or fa­cil­i­ties to re­fine its pre­dic­tive ac­cu­ra­cy and im­prove process out­comes. Se­cure da­ta struc­tures and fed­er­at­ed learn­ing ap­proach­es al­low dig­i­tal twins to "talk" to each oth­er, ac­cel­er­at­ing learn­ing while pre­serv­ing da­ta in­tegri­ty and com­pli­ance. This am­pli­fies val­ue across the en­ter­prise, turn­ing iso­lat­ed dig­i­tal repli­cas in­to a co­or­di­nat­ed sys­tem for mul­ti-site process ex­cel­lence. Of course, to bring all this po­ten­tial in­to GMP man­u­fac­tur­ing, com­pli­ance can’t be an af­ter­thought.

While the de­vel­op­ment of dig­i­tal twins is be­com­ing more and more preva­lent across man­u­fac­tur­ing sec­tors, GMP en­vi­ron­ments present a high­er lev­el of com­plex­i­ty due to the strin­gent reg­u­la­tions and com­pli­ance de­mands that de­fine the phar­ma­ceu­ti­cal in­dus­try. De­ploy­ing dig­i­tal twins in GMP en­vi­ron­ments de­mands more than tech­ni­cal ca­pa­bil­i­ty, it re­quires full align­ment with reg­u­la­to­ry and op­er­a­tional ex­pec­ta­tions. That’s why Aizon’s plat­form is built with:

• Full trace­abil­i­ty and au­ditabil­i­ty to com­ply with 21 CFR Part 11 and EU An­nex 11.
• Rig­or­ous val­i­da­tion pro­to­cols for mod­els and pre­dic­tions.
• Se­cure ar­chi­tec­ture that meets the high­est stan­dards for da­ta in­tegri­ty and cy­ber­se­cu­ri­ty.
• In­ter­op­er­abil­i­ty with ex­ist­ing sys­tems like MES, LIMS, and ERP.
• Da­ta life­cy­cle man­age­ment aligned with AL­COA+ prin­ci­ples
• AI mod­el life­cy­cle gov­er­nance based on cur­rent best prac­tices from the EMA and FDA.

With this ro­bust foun­da­tion, com­pa­nies can adopt dig­i­tal twins with con­fi­dence, ac­cel­er­at­ing time to val­ue while sim­pli­fy­ing com­pli­ance with glob­al reg­u­la­tions.

Cur­rent phar­ma­ceu­ti­cal reg­u­la­tions man­date that dig­i­tal twins us­ing AI, es­pe­cial­ly those sup­port­ing de­ci­sions re­lat­ed to prod­uct qual­i­ty or pa­tient safe­ty, must in­clude a hu­man in the loop. That means any ac­tion pro­posed or pre­dict­ed by the twin must be re­viewed and ap­proved by qual­i­fied per­son­nel be­fore im­ple­men­ta­tion. This re­quire­ment is root­ed in the pre­vi­ous­ly men­tioned GxP-based prin­ci­ples of ac­count­abil­i­ty, trace­abil­i­ty, and val­i­da­tion, en­sur­ing that AI re­mains a tool sup­port­ing rather than re­plac­ing hu­man judg­ment.

How­ev­er, the reg­u­la­to­ry land­scape is evolv­ing. Ini­tia­tives like the CPV of the Fu­ture and the EMA's re­flec­tion pa­pers on AI in­di­cate a shift to­ward more dy­nam­ic, risk-based frame­works. These would al­low for high­er lev­els of au­ton­o­my in AI sys­tems, pro­vid­ed they are ex­plain­able, val­i­dat­ed, and op­er­ate with­in clear­ly de­fined guardrails. At Aizon, we de­sign dig­i­tal twins that align with to­day’s ex­pec­ta­tions while be­ing ready for to­mor­row’s: ful­ly trace­able, au­ditable, and con­fig­urable for both su­per­vised and au­tonomous and se­mi-au­tonomous use. This en­sures that man­u­fac­tur­ers stay com­pli­ant and at the same time fu­ture-proof their op­er­a­tions.

To dri­ve in­no­va­tion while en­sur­ing ad­her­ence to GMP, Aizon’s work is ground­ed in both sci­en­tif­ic rig­or and re­al-world de­ploy­ment. This in­volves both pub­lish­ing ex­ten­sive­ly on the ap­pli­ca­tion of AI and dig­i­tal twins in reg­u­lat­ed man­u­fac­tur­ing, and con­tribut­ing ac­tive­ly to glob­al work­ing groups shap­ing best prac­tices, in­clud­ing the afore­men­tioned CPV of the Fu­ture ini­tia­tive. Through ac­tive in­volve­ment in such ini­tia­tives and con­tri­bu­tions to in­dus­try guid­ance, Aizon is shap­ing not just the tech­nol­o­gy but al­so the reg­u­la­to­ry stan­dards guid­ing its de­ploy­ment.