FDA HALO & Elsa 4.0: Sponsor Guide to AI Submissions

Purpose and Scope. HALO (Harmonized AI & Lifecycle Operations for Data) is FDA’s internal platform consolidating more than 40 disparate application and submission data sources, systems, and portals across FDA centers. FDA has not published a complete inventory of included records, a cross-product “single source of truth,” or a sponsor-facing HALO submission process (^[8]).

Data Integration and Accessibility. FDA announced that HALO consolidates more than 40 application and submission data sources, systems, and portals across FDA centers. FDA also said it had begun integrating HALO with Elsa so staff can query data and build workflows without manually uploading documents within each chat. FDA has not published HALO’s data model, normalization standards, interface design, cross-application retrieval behavior, or complete record inventory. FDA’s Chief AI Officer said that Elsa would soon become the main entry point to FDA systems and data and that it now sits on top of FDA data (^[8]).

Capabilities and Use Cases. FDA states that HALO’s consolidation is expected to enable broader deployment of AI capabilities within agency operations and that the agency has begun integrating HALO with Elsa. FDA has not published specific HALO analytics, cross-record linkages, dashboards, graphs, or inspection use cases (^[8]).

Technical Features. FDA has not publicly detailed HALO’s architecture, data standards, metadata model, APIs, user interface, or search implementation. The published announcement supports only that HALO consolidates agency data sources, systems, and portals and is being integrated with Elsa for internal FDA workflows (^[8]).

Implications. FDA says that reducing tedious work can help staff focus more on science and make workflows more efficient. The agency expects HALO’s consolidation to enable broader deployment of AI capabilities within its operations. FDA has not published measured effects on review times, retrieval speed, safety-signal detection, decision consistency, or pre-go-live issue resolution for HALO (^[8]).

HALO is an internal FDA platform. FDA has not announced that HALO changes sponsor filing channels, creates a cross-center submission schema, or establishes a sponsor-facing “consolidated eCTD” process. Sponsors should follow the applicable published submission requirements and monitor FDA announcements for future changes. HALO is a foundation for AI-augmented internal FDA operations.

Overview. Elsa (“Electronic aSsessment and LifelOng Assistance”) is FDA’s in-house AI assistant for regulatory review, and Elsa 4.0 (launched May 2026) is its latest major upgrade (^[9]). Elsa is a generative Large Language Model (LLM)–based chatbot, akin to an FDA-specific private ChatGPT, trained on internal guidance and knowledge but not on regulated industry’s proprietary data (^[10]). Elsa was first introduced in June 2025 (^[5]) to accelerate document review and is now fully agency-wide. Elsa 4.0 builds on that foundation with dozens of new features to make it more powerful and interactive.

Key New Features. FDA lists custom agents, document generation, quantitative data analysis and visualization, secure web access, voice-to-text dictation, OCR, enhanced chat flexibility, and optimized search for large document repositories among Elsa 4.0’s features. FDA says that its secure-web-access feature uses refreshed secure web data in responses and that Elsa is not connected to the internet. FDA does not publish the individual user workflows, repository scope, context limits, semantic-search design, or specific document-generation examples described above. Elsa does not train on input data or data submitted by regulated industry (^[8]).

Uses in FDA Operations. FDA’s June 2025 announcement identified Elsa uses including accelerating clinical-protocol reviews, shortening scientific evaluations, identifying high-priority inspection targets, summarizing adverse events, performing label comparisons, and generating code to help develop databases for nonclinical applications. FDA’s May 2026 announcement lists Elsa 4.0 features but does not publish the specific cross-application analysis, inspection-checklist, Form 483, real-time-regulatory-intelligence, or decision-support workflows described above. It also states that secure web access uses refreshed secure web data in responses while Elsa is not connected to the internet (^[11]; FDA announcement on Elsa 4.0 and HALO).

Operator Assurance and Security. FDA states that Elsa 4.0 is built in a FedRAMP High Google Cloud Platform environment and does not train on input data or data submitted by regulated industry. FDA also states that subject-matter experts verify inputs, analytic processes, and output implementation at every stage of Elsa’s AI work process. FDA does not publish a separate validation framework or performance assessment addressing reliability or bias (^[8]).

Comparison with Elsa 1.0. The enhancements over Elsa 1.0 (June 2025) are significant. Elsa 1.0 was already proving itself as a generative “work assistant” that could summarize adverse event reports, perform label comparisons, and draft code to support database development (^[12]). Elsa 4.0 adds automation, analytics, and accessibility. (Table 1 below compares key capabilities of Elsa versions 1.0 vs 4.0.) Key new elements in 4.0 – custom AI agents, charting, voice-to-text – greatly expand the helper roles Elsa can play.

Capability	Elsa 1.0 (Jun 2025)	Elsa 4.0 (May 2026)
Document Summarization (text)	Yes – summarize reports, briefs	Yes – improved contextual understanding, supports longer docs
Adverse Event & Safety Reviews	Yes – automated AE summarization ([13])	Yes (unchanged)
Label/Content Comparison	Yes – compare labelling content across submissions ([13])	Yes – enhanced search for key text
Code/Data Generation	Yes – can generate code for databases ([13])	Yes (in addition supports quantitative analytics)
Custom AI Agents	No – single generic chatbot	Yes – build task-specific AI agents for workflows
Document Generation (writing)	Limited – mostly Q&A	Yes – can draft complete documents or sections (e.g. cover letters)
Quantitative Analysis & Visualization	No – text only	Yes – can create charts/graphs from data tables ([14])
Web Search (secure)	No (no web access)	Yes – search certified web content (FDA sites) ([14])
Voice Dictation (speech-to-text)	No	Yes – converts spoken queries into text
OCR (Image-to-Text)	No	Yes – extracts text from scanned documents/images ([14])
Chat Flexibility	Basic (shorter context)	Advanced (longer chats, better memory)
Underlying Data Access	Yes – but required manual upload of docs to Elsa	Yes – directly queries consolidated HALO data ([15])
FedRAMP/Cloud Security	Yes – FedRAMP High (GovCloud AWS) ([16])	Yes – FedRAMP High (Google Cloud), no industry data train ([17])
Human Review Safeguards	Yes – outputs reviewed by experts ([17])	Yes (each output stage verified by staff ([17]))

Table 1: Comparison of Elsa 1.0 and Elsa 4.0 features (based on FDA announcements (^[14]) (^[18])).

Overall, Elsa 4.0 expands FDA’s internal AI-tool features and is being integrated with HALO. FDA has not characterized Elsa as a multi-agent regulatory co-pilot or published evidence supporting the broader workflow and reviewer-performance claims above (^[8]).

“Consolidated eCTD” is not a published FDA sponsor-submission program or requirement. HALO consolidates more than 40 FDA application and submission data sources, systems, and portals for internal agency use, and FDA is integrating it with its internal AI tool, Elsa. FDA has not stated that HALO changes how sponsors transmit submissions or combines product dossiers into a single sponsor-facing HALO package.

For CDER and CBER, eCTD remains the standard format for the submission types specified in FDA guidance. Current supported versions are eCTD v3.2.2 and v4.0. FDA accepts v4.0 only for new NDA, BLA, ANDA, IND, and master-file applications; forward compatibility for existing v3.2.2 applications remains a future phase. Qualifying eCTD submissions of 10 GB or less must be sent through ESG (^[19]; Submit Using eCTD).

FDA began accepting eCTD v4.0 for new NDA, BLA, ANDA, IND, and master-file applications on September 16, 2024. FDA continues to support v3.2.2 and v4.0 and has said it will give advance notice before beginning eCTD v4.0-only electronic submissions. FDA has not linked eCTD v4.0 adoption to HALO compatibility, announced HALO-specific metadata or formats, or stated that sponsors will submit into a single HALO repository (^[19]; FDA announcement).

FDA’s June 2025 announcement states that Elsa was already being used to identify high-priority inspection targets. FDA’s subsequent HALO announcement describes internal consolidation of application and submission data sources and integration with Elsa, but it does not describe inspection-data linkages, inspector workflows, on-site inspection support, or HALO-based sponsor obligations.

FDA has not published inspection procedures, sponsor-system access, inspector dashboards, or record-format requirements involving HALO or Elsa. Sponsors should maintain records and quality systems in accordance with the requirements applicable to their products and operations; they should not treat HALO compatibility as a published inspection requirement (^[8]).

FDA’s announcements describe Elsa as an internal tool for FDA staff and identify high-priority inspection targets as one existing Elsa use case. They do not describe HALO- or Elsa-assisted on-site inspections, sponsor-system access, sponsor data sharing, or HALO-specific mock-audit practices. Sponsors should continue to prepare for inspections under applicable FDA requirements and should monitor FDA communications for any future procedural changes (^[11]; FDA announcement on Elsa 4.0 and HALO).

FDA has not announced a sponsor migration, HALO filing process, or HALO-specific preparation requirements. The practices below are optional internal digital-quality and submission-readiness measures; they do not constitute HALO or Elsa obligations. Sponsors should follow the requirements applicable to their product and submission type.

1. Form a Cross-Functional Implementation Team

Stakeholders. Assemble a core team including Regulatory Affairs, Quality Assurance, IT/Systems, and Supply Chain/Manufacturing representatives. This mirrors successful transitions like EMA’s CTIS adoption, where sponsors created dedicated project teams to manage the shift. Executive sponsorship is critical to allocate resources promptly.

Assessment & Training. Teams may review FDA’s published eCTD requirements and the requirements applicable to their products. FDA has not published sponsor-specific HALO or Elsa manuals, required sponsor training, or a HALO-related implementation timetable. Organizations that choose to use internal AI tools should establish appropriate governance and human review for those tools independently of FDA’s internal Elsa deployment.

2. Audit Existing Data and Systems

Inventory Submissions. Catalogue all active product submissions (INDs, NDAs, BLAs, ANDAs, supplements, DMFs, biologics license supplements, etc.). Note the format (eCTD 3.2.2 vs v4.0) and location (archived offline, in an eCTD publishing tool, etc.). Identify duplicates or redundant documents.

Data Harmonization. Maintain accurate, consistent internal submission records and use FDA’s currently applicable submission standards. FDA has not published HALO-specific sponsor identifiers, metadata requirements, or requirements to modify a sponsor’s RIM or document-management system (^[19]; FDA announcement).

Quality of Electronic Records. Follow the document, validation, and recordkeeping requirements applicable to the product and submission type. FDA states that Elsa 4.0 can convert scanned documents and images into searchable text, but it has not issued HALO- or Elsa-specific sponsor requirements for scan quality, searchable PDFs, version control, or indexing (^[8]).

Security and Access. Use FDA’s published account and transmission instructions for the applicable sponsor-facing channel. For qualifying CDER/CBER eCTD submissions, this includes ESG registration and testing; for eligible CDRH premarket submissions, use the CDRH Portal. FDA has not published a sponsor HALO connection, registration, or feedback interface (^[7]; CDRH Portal).

3. Upgrade Submission Processes

eCTD Assembly. Use the FDA technical conformance guide, FDA Data Standards Catalog, and other requirements applicable to the submission. FDA currently supports eCTD v3.2.2 and v4.0; v4.0 is supported for new NDA, BLA, ANDA, IND, and master-file applications, and forward compatibility for existing v3.2.2 applications is not yet available. Follow FDA notices for implementation changes. Local validation may support compliance with the applicable eCTD requirements, but FDA has not announced that sponsors submit validated packages to HALO or must align with HALO-specific expectations.

Submission Quality. Prepare submissions according to the applicable FDA technical and validation requirements. FDA has not published HALO- or Elsa-specific sponsor formatting, tagging, OCR, or executive-summary requirements (^[7]; FDA announcement).

Metadata and Audit Trails. Follow the metadata, validation, and recordkeeping requirements that apply to the submission and product. FDA has not published a HALO lifecycle-tracking requirement, sponsor audit-trail interface, or requirement for metadata intended to help Elsa find context (^[7]; FDA announcement).

Communication with FDA. Monitor FDA announcements, Federal Register notices, and the product- and center-specific submission resources that apply to the submission. FDA has not announced a HALO sponsor implementation timetable or a sponsor transition period; organizations should not infer one from other regulators’ programs.

4. Integrate and Leverage AI Tools

Internal AI Adoption. Just as FDA uses Elsa, sponsors should pilot AI tools for drafting and review. Tools like RIM-integrated GenAI assistants can automate parts of dossier creation. Industry reports indicate early AI pilots extracted regulatory data fields with roughly 90% accuracy while processing up to 80% faster (^[20]). Sponsors should harness these to accelerate indexing of existing data (building “knowledge bases” from past submissions) and to pre-screen for consistency.

Elsa Familiarization. Elsa is an internal FDA tool. FDA has not directed sponsors to emulate Elsa, prepare documents for Elsa, or use any particular AI system. Sponsors should prepare submissions in the format and with the content required by the applicable FDA requirements; any internal AI use should be governed under the organization’s own policies and controls.

5. Inspection Readiness and Quality Systems

Inspection Readiness and Quality Systems. Maintain inspection records, quality systems, and document control in accordance with requirements applicable to the product and operation. FDA has not announced that sponsors must digitize records for HALO, make audit trails queryable by HALO, or simulate HALO- or Elsa-assisted inspections (^[8]).

Labeling and Safety Monitoring. Maintain labeling and safety reporting in accordance with the requirements applicable to the product and submission type. FDA has not published a process for sponsors to register safety databases in HALO or to provide Elsa access to sponsor systems.

6. Collaboration and Continuous Improvement

Industry Forums and Pilot Programs. Organizations may monitor FDA announcements and participate in any FDA programs that are formally announced. FDA has not announced a HALO sponsor pilot, sponsor working group, or HALO requirements against which sponsors can test internal tools.

Monitor Regulatory Feedback. Monitor FDA announcements and follow published submission, review, and correspondence procedures. FDA has not announced sponsor filing through HALO or a process under which sponsors receive feedback about Elsa’s interpretation of submitted data (^[8]; Submit Using eCTD).

Future Updates. FDA may publish additional information about HALO or Elsa. Sponsors should monitor FDA communications and update their processes when FDA publishes a sponsor-facing requirement or implementation notice.

To ground this playbook, we examine data and case examples from FDA and industry:

• Submission Volumes: FDA’s ESG statistics list 319,828 CDER, 97,133 CBER, and 4,664,504 CDRH submissions for 2025. These are ESG reporting categories; FDA does not characterize the 4,664,504 CDRH figure as largely eSTAR or 510(k) premarket submissions. Separately, CDRH’s 2025 annual report states that it received 21,780 medical-device submissions (^[21]; CDRH Annual Report 2025).

• FDA Review Efficiency: FDA has stated that Elsa is intended to help employees work more efficiently and has identified several operational uses. FDA has not published a HALO- or Elsa-attributable change in PDUFA review times, review duration, or data-retrieval speed; no numerical effect should be inferred (^[11]; FDA announcement on Elsa 4.0 and HALO).

• Industry Projected Savings: According to Contract Pharma interviews, current AI pilots achieved dramatic productivity gains, with more than a dozen fields of data extracted with around 90% accuracy and up to 80% faster processing (^[20]). These sources also note that even 1–2 hours saved per submission (out of hundreds) yields massive ROI. While these figures are vendor-reported, they are echoed by industry executives. Similarly, a regulatory-industry survey found three in five (60%) regulatory leaders reporting that their workload has grown far faster than normal company growth, fueling AI adoption (^[22]). These insights reinforce that the HALO/Elsa shift is timely: sponsors have been struggling with data overload and welcome automation.

• Clinical-trial submission context: EU CTIS is a separate sponsor-facing system and does not establish a HALO filing model. FDA has not announced sponsor access to HALO or a HALO pilot-submission process. Sponsors should use the currently applicable FDA submission channels and monitor FDA announcements for any future changes (^[8]; Submit Using eCTD).

• Case Example – Regulatory AI Tools: Vendors like DossiAIr and Cruxi highlight how private-sector platforms can auto-generate submissions. DossiAIr markets an AI-assisted "Plan, Build, Validate, Submit" workflow that it claims can compress dossier preparation from three to six months down to two to three weeks (^[23]). Cruxi's AI-assisted 510(k) platform has drawn customer praise: one client reported the workflow "saved us thousands of dollars and time and significantly enhanced the quality of submission" (^[24]), while another called a generated report "highly insightful" (^[25]). These case studies underscore that automated submission technology is maturing. Sponsors may partner with such tools or apply similar techniques internally to prepare dossiers more efficiently while meeting currently applicable FDA submission requirements.

• Quality and Data Integration: FDA states that Elsa 4.0 can convert scanned documents and images into searchable text and offers optimized search for large document repositories. FDA has not published OCR trial results, HALO-specific document-legibility requirements, or requirements for sponsors to prepare printed records for HALO accessibility.

Overall, FDA’s published materials describe internal HALO and Elsa capabilities, not a sponsor migration program. Sponsors should follow currently applicable FDA requirements and monitor FDA communications for any future sponsor-facing changes.

While FDA’s HALO/Elsa is unprecedented domestically, similar initiatives inform our analysis:

• EMA’s CTIS Rollout (EU). The Clinical Trials Information System (CTIS), launched under the 2022 EU Clinical Trials Regulation, consolidated trial submissions across EU member states. From January 31, 2023, all new clinical trials in the EU had to use CTIS as a single portal (^[26]). The EMA reported resolving ~80% of critical issues before this go-live by intense preparation with member states. Sponsors faced vesting: multinational trial sponsors had to migrate ongoing trials from old national systems into CTIS. Lessons learned included establishing CTIS centre accounts early, double-checking translations and coding in the application, and engaging national authorities proactively. HALO is an FDA internal platform, whereas CTIS is an EU sponsor-facing clinical-trial system. The CTIS rollout does not establish a HALO filing model or a HALO sponsor transition expectation. Sponsors should monitor FDA communications for formally announced sponsor-facing changes.

• Electronic device submissions. eSTAR is not an upcoming requirement. Unless exempted, FDA has required electronic 510(k) submissions using eSTAR since October 1, 2023, and electronic De Novo submissions using eSTAR since October 1, 2025. These CDRH requirements are distinct from CDER/CBER eCTD requirements, and FDA has not announced a HALO-based common document across drug and device submissions (^[27]).

• Regulatory AI Adoption Trends. Several industry publications emphasize that companies are rapidly adopting AI. A survey covered by European Pharmaceutical Manufacturer reported that 35% of senior regulatory professionals already use some form of AI for regulatory work and a further 42% plan to invest within 18 months, driven by unrelenting workload increases (^[28]). Many respondents pointed out that outdated IT (45%), perceived risk exceeding benefit (44%), and insufficient data quality (42%) slow adoption (^[29]). Nearly half (48%) of respondents expect AI to transform routine regulatory work within the same period (^[30]). These survey perspectives concern private-sector AI adoption; they do not establish a HALO preparation requirement or competitive consequence for FDA sponsors.

• FDA’s Initial Elsa Deployment. FDA reported that it launched Elsa agency-wide ahead of schedule and under budget. FDA described Elsa as a tool intended to help employees work more efficiently and listed operational uses, but it did not publish reviewer-performance results or direct sponsors to adopt comparable AI tools (^[11]).

• HALO scope: FDA has announced internal consolidation of application and submission data sources, systems, and portals, but has not published linked sponsor product profiles, an IND-to-NDA reconciliation workflow, or an Elsa inconsistency-flagging procedure. Sponsors should follow the applicable published submission and amendment requirements (^[8]; Submit Using eCTD).

These examples distinguish sponsor-facing systems such as EU CTIS and CDRH electronic submission processes from FDA’s internal HALO and Elsa initiatives. They do not establish a HALO filing model, sponsor transition requirement, or competitive consequence. Sponsors should follow the FDA requirements applicable to their products and monitor FDA for any future sponsor-facing announcement.

The HALO platform and Elsa 4.0 mark a watershed in regulatory science, with far-reaching implications:

Operational Implications. FDA describes HALO and Elsa as internal modernization efforts intended to reduce tedious work and support agency operations. FDA has not published quantified review or inspection effects, HALO-specific sponsor formats, sponsor training, or a sponsor system-upgrade requirement. Sponsors should continue to use the applicable current submission channels and formats unless FDA announces a sponsor-facing change (^[8]; FDA eCTD information).

Data Quality and Integrity. HALO’s power depends on high-quality data. The “garbage in, garbage out” maxim applies: poorly organized or inaccurate submissions can lead Elsa to inaccurate inferences. Sponsors must therefore audit and clean data meticulously. This concern is echoed by industry surveys: 42% of respondents flagged insufficient data quality or availability as a barrier to AI adoption (^[31]). As a countermeasure, regulatory plans should incorporate data governance – ensuring document accuracy, proper version control, and metadata fidelity. Organizations may even employ AI itself to identify inconsistencies pre-submission (the same way Elsa does post-submission).

Human Oversight. FDA states that subject-matter experts verify inputs, analytic processes, and output implementation throughout Elsa’s AI work process. The agency’s announcement does not characterize Elsa outputs as “advisory” or publish a general validation framework. Organizations using their own AI tools should apply governance and human-review controls appropriate to their intended use and applicable requirements (^[8]).

Regulatory Science Evolution. FDA has described HALO and Elsa as internal modernization efforts and has not announced sponsor-facing HALO requirements, an AI-submission standard, or the named sponsor programs. Sponsors should monitor FDA’s official announcements and follow the submission, product, and quality requirements currently applicable to their products (^[8]).

Global Harmonization Pressure. Other regulators will feel pressure to match FDA. As mentioned, China’s NMPA has an “AI+ drug regulation” roadmap aiming for intelligent oversight by 2035 (^[32]). The UK’s MHRA published its AI regulatory strategy for medical products, setting out four key principles (safety, security and robustness; transparency and explainability; fairness, accountability and governance; and contestability and redress) to guide safe, explainable AI in regulation through 2030 (^[33]). The International Council for Harmonisation (ICH) or WHO may accelerate guidance on AI in submissions. Sponsors operating globally will need to juggle multiple AI-assisted systems. One potential future is cross-national convergence: a world where agency portals interoperate. HALO could seed such harmonization if FDA opens APIs or data exchange standards. This might lead, for instance, to joint inspections where US and EU authorities share HALO/CTIS findings, boosting multinational consistency.

Cybersecurity and Privacy. FDA states that Elsa is built in a FedRAMP High Google Cloud Platform environment and does not train on input data or data submitted by regulated industry. FDA has not announced sponsor connections, APIs, or network interfaces to HALO. Sponsors should follow the security and transmission requirements for the sponsor-facing FDA channels that apply to their submissions and monitor FDA communications for future changes (^[8]).

Sponsor impact. HALO and Elsa are internal FDA initiatives. FDA has not announced HALO-specific sponsor preparation requirements, a sponsor migration timetable, new filing channels, inspection procedures, or evidence that a sponsor’s internal technology choices affect review timing. Sponsors should continue to use the applicable FDA formats and channels unless FDA publishes a sponsor-facing change.

In summary, HALO and Elsa expand FDA’s internal AI and data-platform capabilities. Sponsors should continue to comply with currently applicable FDA submission and product requirements and monitor FDA communications for any future sponsor-facing changes.

The FDA’s May 2026 announcement of HALO and Elsa 4.0 expands AI-augmented internal review operations. HALO consolidates FDA data sources, systems, and portals, while Elsa gives FDA staff new AI capabilities to query data and build workflows without manually uploading documents to each chat. For industry sponsors, the immediate obligation is to maintain compliance with the applicable published submission channel and format; FDA has not announced a requirement to interface directly with HALO or Elsa (^[8]; Submit Using eCTD).

This report distinguishes FDA’s published internal HALO and Elsa activities from sponsor-facing requirements. The optional practices discussed above do not replace applicable FDA submission, quality, inspection, or product requirements.

FDA may provide further information as its internal systems evolve. Until FDA announces a sponsor-facing change, sponsors should use the currently applicable submission channels and formats and avoid treating HALO or Elsa as a new filing or inspection obligation.

References: Key sources include the FDA’s official press release and related news coverage (^[15]) (^[1]), FDA guidance on eCTD submissions (^[4]), industry analyses on AI in regulatory affairs (^[20]) (^[6]), and international regulatory agency publications (^[34]) (^[26]). The cited sources include primary FDA materials as well as clearly identified secondary and vendor materials; sponsor obligations should be determined from the applicable FDA requirements and notices.