AI, Gene Therapy & Radioligand Theranostics
Weekly Update Digest | February 23 – March 1, 2026 (UTC)
Welcome to this week's digest. This edition covers the latest official announcements, regulatory actions, and expert perspectives from the frontlines of artificial intelligence, gene therapy, and radioligand theranostics. All links lead directly to source material from regulatory agencies, peer-reviewed journals, clinical trial registries, and official press releases.
🤖 Artificial Intelligence
Connecticut Attorney General Issues Detailed Memo on Applying Existing State Laws to AI
The Connecticut Attorney General released a comprehensive memorandum explaining how existing state law (civil rights, privacy/data security, unfair trade practices, and antitrust) already applies to AI systems and their deployers, without waiting for AI-specific statutes. It explicitly flags uses such as deepfakes, fake reviews, AI-enabled scams, and algorithmic collusion in pricing as potential violations under current law, and reminds developers and operators that they can be liable even when using third-party models or tools.
Why it matters: For anyone deploying AI into consumer- or patient-facing workflows, the message is that "we used an algorithm" is not a legal defense; the same old statutes will be used aggressively against new AI behaviors. Expect other state AGs to copy-paste this playbook, increasing litigation and enforcement risk around opaque or poorly governed AI use. In other words: if your compliance team is still treating AI as "R&D only", this memo is a polite (but legally sharp) reality check.
Source: Office of the Connecticut Attorney General →🧬 Gene Therapy
FDA Issues Draft "Plausible Mechanism" Framework to Accelerate Individualized Gene and RNA Therapies
FDA has released draft guidance introducing a "Plausible Mechanism Framework" for individualized therapies targeting specific genetic, cellular, or molecular abnormalities in ultra-rare, severely debilitating or life-threatening conditions, where conventional randomized trials are often infeasible. The framework is aimed squarely at genome editing products and RNA-targeted therapies such as antisense oligonucleotides, and sets out conditions under which a single adequate and well-controlled study plus robust confirmatory evidence can support approval.
Why it matters: Sponsors must: (1) identify a well-characterized disease-causing abnormality, (2) show the therapy directly targets this biology, (3) leverage detailed natural history in untreated patients, (4) confirm target engagement or successful editing, and (5) demonstrate meaningful clinical benefit or validated surrogate response. For gene and cell therapy developers, this is a structural shift: FDA is not lowering the statutory standard for "substantial evidence", but is formalizing a route to reach it with far fewer patients—provided the biology is airtight and the data package is mechanistically rigorous.
Source: FDA / HHS →☢️ Radioligand Theranostics
Real-World Data Support Earlier Pluvicto Use in Taxane-Naïve PSMA-Positive mCRPC
Novartis reported US real-world evidence from its PRECISION platform showing Pluvicto (lutetium Lu-177 vipivotide tetraxetan) achieved a median PFS of 13.5 months (95% CI 11.7–14.7) in taxane-naïve PSMA-positive mCRPC patients previously treated with ≥1 ARPI. When Pluvicto was used after only one ARPI, median PFS was 15.8 months versus 12.7 months after multiple ARPIs, supporting earlier RLT introduction rather than exhausting sequential ARPIs. A second analysis indicated that subsequent systemic therapies remained clinically active after Pluvicto discontinuation.
Why it matters: The PFS estimates are consistent with PSMAfore trial results that underpinned earlier-line approvals, adding external validity in more heterogeneous real-world practice. For theranostic programs and for clinical ops planning, this is an important signal that sequencing RLT after first ARPI—and not as a salvage of last resort—may be both clinically rational and increasingly expectations-setting for payers and clinicians.
Source: Novartis (ASCO GU 2026) →📡 Expert Signals
Federal Reserve Board: Operationalizing AI
In a speech on "Operationalizing AI at the Federal Reserve", Governor Christopher J. Waller described how the Fed is embedding AI into internal operations (text analysis, sentiment extraction, workflow tools), while emphasizing that a central bank cannot "move fast and break things". He stresses strict model risk management, information-security controls, human accountability, and ongoing monitoring.
Insight: The take-home for any regulated sector (including pharma) is that supervisors now have a live, public example of what "responsible AI adoption by a systemically important institution" looks like: aggressive use of AI internally, but wrapped in governance, validation, and documentation that would make most tech startups cry.
Source: Federal Reserve Board →Nature Biotechnology: FDA Clears "Cellular Rejuvenation" Gene Therapy Trial
Nature reports that FDA has cleared Life Biosciences to begin a phase 1 trial of ER-100, an AAV-based gene therapy delivering three Yamanaka factors (OCT4, SOX2, KLF4) to partially reprogram retinal cells in an eye disease, aiming to reverse age-associated damage without full dedifferentiation. This is the first human trial explicitly framed as "cellular rejuvenation" via partial epigenetic reprogramming.
Insight: The piece underscores both the transformative potential of restoring youthful epigenetic states and the substantial safety concerns around tumorigenicity. It is an early but clear signal that regulators are willing to entertain therapy concepts that would have sounded like science fiction a decade ago—as long as the vector, targeting, and control logic are rigorously justified.
Source: Nature Biotechnology →Nature: AI is Threatening Science Jobs—Which Ones Are Most at Risk?
Nature highlights that AI is already eroding demand for some research roles centered on data analysis, coding, and routine modeling, while wet-lab experimentalists and clinicians remain relatively insulated in the near term. The article notes that tasks involving summarization, literature scanning, drafting, and standard data workflows are highly automatable.
Insight: This shifts the value proposition toward experimental design, complex interpretation, cross-disciplinary integration, and hands-on work that cannot be offloaded to models easily. For research organizations, this is a not-so-subtle nudge to redesign teams and training around complementary human–AI skill mixes.
Source: Nature →📰 In Case You Missed It
FDA Outlines More Flexible CMC Expectations for Cell and Gene Therapies
In January, FDA announced a more flexible, risk-based approach to chemistry, manufacturing, and controls (CMC) for CGTs, including phased CMC data expectations, flexibility around traditional "three PPQ lots", and tolerance for certain manufacturing changes supported by comparability data.
Why still relevant: This is directly synergistic with the new Plausible Mechanism framework: sponsors now have clearer room to stage CMC investments for individualized or ultra-rare programs without over-engineering early, while still maintaining robust quality and safety expectations.
Source: FDA →MHRA Approves Earlier-Line Pluvicto Indication in PSMA-Positive mCRPC
On 3 February 2026, the UK MHRA authorized Pluvicto for PSMA-positive mCRPC patients who have progressed on ARPI therapy and are considered appropriate to delay taxane-based chemotherapy, effectively moving RLT earlier in the treatment pathway. The decision is based on PSMAfore phase III data showing a 59% reduction in risk of radiographic progression or death versus ARPI switch.
Why still relevant: In combination with the new real-world data, this shapes payer and clinician expectations that RLT is not just a "post-everything" salvage but a legitimate pre-taxane option—relevant for both future trial designs and health-economic modeling.
Source: Novartis UK →U.S. Treasury Issues AI Lexicon and Financial Services AI Risk Management Framework
Treasury released a shared AI Lexicon and a Financial Services AI Risk Management Framework (FS AI RMF) tailored to the financial sector, aligning with NIST's AI Risk Management Framework but operationalized for banks and insurers. The documents provide structured guidance on AI governance, data practices, model risk, explainability, fraud, and identity.
Why still relevant: While sector-specific, they are de facto reference material for any institution trying to convince regulators that its AI risk controls are more sophisticated than "we had a meeting about it once".
Source: U.S. Treasury →Nature: AI Tools Can Now Design Genomes and Fully Synthetic Viruses
A Nature News article describes how AI-assisted "generative biology" is being used to design new proteins, artificial genes expressible in mammalian cells, and even an entirely synthetic virus, marking a qualitative shift from analysis to design of biological systems.
Why still relevant: The piece emphasizes both the enormous upside (rapid vaccine and therapeutic design) and the dual-use/biosafety concerns. For any organization working at the interface of AI and biology, this is the backdrop against which regulators will increasingly assess your threat-modeling and safeguards.
Source: Nature →Nature: FDA Go-Ahead for First Cellular Rejuvenation Gene Therapy Trial in Humans
The Life Biosciences ER-100 program signals that regulators are prepared to evaluate interventions aimed at "reversing" cellular age, not just treating monogenic deficits. This has implications for future indications (neurodegeneration, systemic aging) and for how risk–benefit is judged when the therapeutic goal edges from "disease treatment" toward "rejuvenation".
Why still relevant: Clinical development teams considering anything in that gray zone should be studying this case closely rather than just name-dropping Yamanaka factors in pitch decks.
Source: Nature →🎯 Practical Implications
- Compliance & Liability: Sponsors using AI in trial recruitment, imaging workflows, safety signal detection, or patient-facing apps should map each AI use-case to existing privacy, consumer protection, and antitrust laws, assuming enforcement will follow the Connecticut/California AG model: third-party models and "black-box" vendors do not shield you from liability.
- Rare Disease Strategy: For ultra-rare gene and RNA therapies, development strategies should be re-tooled around the Plausible Mechanism framework: build a watertight mechanistic package (target biology, editing/ASO pharmacology), invest early in natural history and external controls, and pre-specify validated biomarkers or clinical endpoints.
- CMC Flexibility: Cell and gene therapy CMC plans should explicitly leverage FDA's new flexibility: stage process validation and analytical maturity in line with clinical phase, use comparability-driven change management instead of freezing early processes, but still design robust potency/identity/safety assays that can scale to licensure.
- Theranostics Sequencing: In radioligand theranostics, Pluvicto's real-world PFS in chemo-naïve mCRPC and maintained activity of post-RLT systemic therapies support trial and guideline strategies that place RLT after first ARPI rather than after multiple ARPIs and chemotherapy; sequencing models should explicitly test "earlier RLT + preserved downstream options" scenarios.
- AI Governance: For internal AI adoption (e.g., protocol drafting assistants, risk-based monitoring analytics), copying the Federal Reserve's approach is safer than copying Silicon Valley: embed AI into workflows where it augments staff, wrap it in formal model risk management processes, and ensure humans remain clearly accountable for decisions.