Share with your CDO
Kallik, whose enterprise labelling software runs at Kenvue, Cardinal Health, and Procter and Gamble, is pushing a pointed argument into a market that doesn’t want to hear it: pharma manufacturers deploying generative AI over unstructured legacy data aren’t accelerating compliance, they’re building a compliance timebomb. CEO Gurdip Singh frames the fix as a data-first migration workflow, anchored in Kallik’s own AToM tool and Veraciti platform, before any LLM integration begins. Gartner’s parallel prediction, that most AI projects at data-unready organizations will be abandoned, gives the warning independent cover.
What this means for your business
The CDOs most exposed here are the ones who’ve already said yes to an AI pilot in a regulated workflow without first auditing what that AI will actually read. Pharma labelling isn’t abstract, it’s the specific text, font, date format, and market-specific regulatory language on a physical product. An LLM querying fragmented source records doesn’t hallucinate in a vacuum; it produces a label that clears an internal review, ships, and then fails a regulatory inspection in a market where the date format reads differently. That’s not a data quality problem, it’s a recall.
Kallik is a vendor with a product to sell, which means their “data-first” prescription conveniently routes through their own migration tooling, and that incentive likely inflates how broken the average legacy environment actually is. But the underlying structural claim holds independent of that tilt. Autonomous agents, meaning AI systems that execute multi-step tasks without human sign-off at each step, are only as reliable as the records they reason over. In a domain where every output is a regulated artifact, the failure mode isn’t a wrong answer on a dashboard, it’s a compliance event with legal exposure. The governance architecture has to precede the automation, full stop.
The budget decision this reframes isn’t the AI line item, it’s the data migration project that keeps getting deferred. If your organization is carrying a legacy labelling or content repository that nobody wants to pay to clean up, and an AI initiative is now downstream of it, that deferral just became a liability rather than a backlog item. The falsification condition is simple: if a manufacturer can show audit-ready, version-controlled, validated source data already feeding their AI pipeline, Kallik’s warning doesn’t apply to them. Most can’t show that.
Based on reporting from Kallik warns pharma manufacturers against AI ‘compliance timebomb’, originally published 2026-07-16 10:11:00.
