<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>Pharmasight — AI in pharma, built to survive the audit.</title>
    <link>https://www.pharmasight.tech/</link>
    <atom:link href="https://www.pharmasight.tech/feed.xml" rel="self" type="application/rss+xml"/>
    <description>Pharmasight is an AI consultancy for pharmaceutical and life-sciences companies. We design, validate and operate AI systems across discovery, clinical, regulatory, manufacturing, commercial and post-market — inside GxP.</description>
    <language>en-gb</language>
    <lastBuildDate>Sun, 30 Aug 2026 09:00:00 GMT</lastBuildDate>
    <item>
      <title>How to Validate a Non-Deterministic LLM Without Pretending It Is Deterministic</title>
      <link>https://www.pharmasight.tech/blog/validate-llm-gxp-non-deterministic/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/validate-llm-gxp-non-deterministic/</guid>
      <description>A test strategy for generative AI in GxP: qualify the deterministic guardrail layer, set abstention thresholds you can justify, and define revalidation triggers that actually fire.</description>
      <category>GxP, validation &amp; assurance</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Named Data Ownership Beats Any Agent</title>
      <link>https://www.pharmasight.tech/blog/rim-idmp-data-ownership/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/rim-idmp-data-ownership/</guid>
      <description>Only 1 of 59 organisations is future-ready for RIM. The 2026 benchmark says named data accountability, not AI, predicts regulatory operations performance.</description>
      <category>Regulatory affairs</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Review by Exception After Twenty Years of Trying</title>
      <link>https://www.pharmasight.tech/blog/review-by-exception-electronic-batch-records/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/review-by-exception-electronic-batch-records/</guid>
      <description>Pfizer cut EBR implementation cost per site by 80% and cycle times by over 11% without changing MES vendors. What actually moved, and what the regulations permit.</description>
      <category>Manufacturing, CMC &amp; supply</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Your Validated Model Has a Retirement Date</title>
      <link>https://www.pharmasight.tech/blog/model-version-control-gxp-change-control/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/model-version-control-gxp-change-control/</guid>
      <description>One deployment property decides whether a validated model is swapped underneath you — and what binding Annex 11 clause 10 actually requires when it happens.</description>
      <category>GxP, validation &amp; assurance</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Which GCP Clause Actually Governs Your Clinical AI Model</title>
      <link>https://www.pharmasight.tech/blog/ich-e6-r3-ai-governance-clauses/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/ich-e6-r3-ai-governance-clauses/</guid>
      <description>ICH E6(R3) never names AI. The clauses that actually govern a clinical AI model, what binds on 30 August 2026 and what is still only draft.</description>
      <category>Clinical development</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Human Oversight Is a Control You Qualify, Not a Sentence in Your Risk Assessment</title>
      <link>https://www.pharmasight.tech/blog/human-in-the-loop-qualification-gxp/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/human-in-the-loop-qualification-gxp/</guid>
      <description>Draft EU GMP Annex 22 makes human-in-the-loop a monitored manual process, not a risk-assessment sentence. What binds you now, what lapses, and the numbers to keep.</description>
      <category>GxP, validation &amp; assurance</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Costing a Health Authority Query Response Before You Automate It</title>
      <link>https://www.pharmasight.tech/blog/health-authority-query-response-automation-economics/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/health-authority-query-response-automation-economics/</guid>
      <description>The only public HAQ benchmark surveyed 12 companies in 2022. Here is how to build a measured baseline instead, and what 200 hours per query really costs.</description>
      <category>Regulatory affairs</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Validating AI Under GxP in 2026: What Binds You, What Is Draft, What Inspectors Cite</title>
      <link>https://www.pharmasight.tech/blog/gxp-ai-validation/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/gxp-ai-validation/</guid>
      <description>What actually binds AI systems under GxP on 30 August 2026: the 2011 Annex 11, 21 CFR 211.22(c), the EU AI Act, and which AI texts are still only drafts.</description>
      <category>GxP, validation &amp; assurance</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>FDA's First AI Warning Letter Cites a 1978 Rule</title>
      <link>https://www.pharmasight.tech/blog/fda-first-ai-warning-letter-purolea/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/fda-first-ai-warning-letter-purolea/</guid>
      <description>FDA Warning Letter 320-26-58 to Purolea, dated 2 April 2026, cites 21 CFR 211.22(c) for unreviewed AI-generated GMP records. What it means for AI governance.</description>
      <category>Manufacturing, CMC &amp; supply</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Your FDA Reviewer Switched Models</title>
      <link>https://www.pharmasight.tech/blog/fda-elsa-ai-model-switch-sponsors/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/fda-elsa-ai-model-switch-sponsors/</guid>
      <description>FDA's Elsa moved off Claude in early 2026 under a federal directive. What a regulator-side model swap does to review consistency, and what sponsors should do.</description>
      <category>Regulatory affairs</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>FDA's Seven-Step AI Credibility Framework — And the Regulatory Operations AI It Never Covered</title>
      <link>https://www.pharmasight.tech/blog/fda-ai-credibility-framework-scope/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/fda-ai-credibility-framework-scope/</guid>
      <description>What FDA's 7-step AI credibility assessment framework covers, what its scope section excludes, and which rules actually bind eCTD publishing and RIM AI.</description>
      <category>Regulatory affairs</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>What EMA Permits When AI Drafts Your Product Information</title>
      <link>https://www.pharmasight.tech/blog/ema-ai-drafted-smpc-labelling-epi/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/ema-ai-drafted-smpc-labelling-epi/</guid>
      <description>EMA section 2.3.5 permits AI-drafted SmPC, labelling and leaflets under close human supervision and a factual and syntactic quality review. What that takes.</description>
      <category>Regulatory affairs</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>eCTD 4.0 in Japan, Europe and America — and Why Publishing Should Stay Deterministic</title>
      <link>https://www.pharmasight.tech/blog/ectd-v4-timelines-publishing-automation/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/ectd-v4-timelines-publishing-automation/</guid>
      <description>PMDA moved to eCTD v4.0 in April 2026, EMA opened optional use in December 2025, FDA has set no date. What that means for publishing automation.</description>
      <category>Regulatory affairs</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Serialisation Proves Where a Package Went, Not Whether the Next One Ships</title>
      <link>https://www.pharmasight.tech/blog/dscsa-data-drug-shortage-prediction/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/dscsa-data-drug-shortage-prediction/</guid>
      <description>DSCSA serialisation data records custody, not capacity. Why shortage-prediction models miss, what EPCIS exception management is genuinely good for, and the 2026 dates that moved.</description>
      <category>Manufacturing, CMC &amp; supply</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Digital Twins Cannot Release Your Batch</title>
      <link>https://www.pharmasight.tech/blog/digital-twin-real-time-release-211-110/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/digital-twin-real-time-release-211-110/</guid>
      <description>FDA's January 2025 draft on 21 CFR 211.110 permits in-line measurement but rules out control strategies relying solely on process models.</description>
      <category>Manufacturing, CMC &amp; supply</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Computer Software Assurance Is Device Guidance — Cite It for a Drug System and Fail</title>
      <link>https://www.pharmasight.tech/blog/csa-vs-csv-drug-sponsors/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/csa-vs-csv-drug-sponsors/</guid>
      <description>FDA's February 2026 CSA guidance covers device production and QMS software under 21 CFR 820. What binds drug sponsors instead, and when the boundary moves.</description>
      <category>GxP, validation &amp; assurance</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>EU GMP Annex 22, Clause by Clause</title>
      <link>https://www.pharmasight.tech/blog/annex-22-requirements-clause-by-clause/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/annex-22-requirements-clause-by-clause/</guid>
      <description>A clause-by-clause reading of draft EU GMP Annex 22: acceptance criteria, test data, independency, explainability and operation, and what each costs.</description>
      <category>GxP, validation &amp; assurance</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>EU GMP Annex 22: What the AI Annex Actually Says</title>
      <link>https://www.pharmasight.tech/blog/annex-22-artificial-intelligence-gmp/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/annex-22-artificial-intelligence-gmp/</guid>
      <description>Draft EU GMP Annex 22 permits only static, deterministic AI models in critical GMP applications and excludes generative AI. What it requires, its status in 2026, and what to do now.</description>
      <category>GxP, validation &amp; assurance</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Nineteen Pages, Seventeen Chapters: The Annex 11 Revision Clauses That Cost Money</title>
      <link>https://www.pharmasight.tech/blog/annex-11-revision-cost-clauses/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/annex-11-revision-cost-clauses/</guid>
      <description>The July 2025 draft Annex 11 runs to 19 pages and 17 sections. Clauses 7.5, 12.6, 12.8 and 9.8 are the ones with a budget attached.</description>
      <category>GxP, validation &amp; assurance</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Nine Contract Clauses Your AI Vendor Will Refuse to Sign</title>
      <link>https://www.pharmasight.tech/blog/ai-vendor-qualification-annex-11-contracts/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/ai-vendor-qualification-annex-11-contracts/</guid>
      <description>Draft EU GMP Annex 11 clause 7.5 sets out nine contract contents for GxP service providers. Which ones a hosted LLM API can meet in 2026, and which it cannot.</description>
      <category>GxP, validation &amp; assurance</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Ninety-Seven Percent Faster, Seventy Percent Complete</title>
      <link>https://www.pharmasight.tech/blog/ai-regulatory-writing-autoind-quality/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/ai-regulatory-writing-autoind-quality/</guid>
      <description>The Weave–Takeda AutoIND preprint: 18,870 pages drafted in 3.7 hours, 69.58% quality, emphasis at 6.7%, and a 100% GLP omission rate.</description>
      <category>Regulatory affairs</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>What Regulators Have Actually Approved, Cited and Refused in AI-Enabled Manufacturing</title>
      <link>https://www.pharmasight.tech/blog/ai-pharmaceutical-manufacturing-regulatory-reality/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/ai-pharmaceutical-manufacturing-regulatory-reality/</guid>
      <description>FDA has accepted 191 emerging technology proposals since 2014 and just 9 involve AI or modelling. What regulators have approved, cited and refused, with dates.</description>
      <category>Manufacturing, CMC &amp; supply</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Drift, Abstention and Audit Trails: The Operating Plan a Deployed GxP Model Needs</title>
      <link>https://www.pharmasight.tech/blog/ai-drift-monitoring-revalidation-gxp/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/ai-drift-monitoring-revalidation-gxp/</guid>
      <description>Drift monitoring, abstention rates and audit trails for a deployed GxP model: the 6 signals, the metrics, and when a retest becomes mandatory.</description>
      <category>GxP, validation &amp; assurance</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Why Adding Agents Makes Deviation Investigations Worse</title>
      <link>https://www.pharmasight.tech/blog/ai-deviation-investigation-multi-agent-risk/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/ai-deviation-investigation-multi-agent-risk/</guid>
      <description>Controlled studies show multi-agent AI degrades sequential reasoning 39-70%. What that means for GMP deviation investigation, CAPA root cause and batch release.</description>
      <category>Manufacturing, CMC &amp; supply</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
    <item>
      <title>Every Disclosed AI Cycle-Time Number in Clinical Development, and What Each Measures</title>
      <link>https://www.pharmasight.tech/blog/ai-clinical-development-cycle-time-evidence/</link>
      <guid isPermaLink="true">https://www.pharmasight.tech/blog/ai-clinical-development-cycle-time-evidence/</guid>
      <description>Novo Nordisk 90%, AutoIND 97%, AstraZeneca 85%, Novartis 83-87%: what each AI clinical development number actually timed, and why none of them is time to approval.</description>
      <category>Clinical development</category>
      <dc:creator>Tareef Roustom</dc:creator>
      <pubDate>Sun, 30 Aug 2026 09:00:00 GMT</pubDate>
    </item>
  </channel>
</rss>
