Know Where You Stand

The Challenge

The company's AI-driven credit scoring model fell squarely under EU AI Act Annex III as a high-risk system — triggering Articles 9–15 obligations — while simultaneously facing GDPR Article 22 restrictions on automated decision-making. Their legal team was manually cross-referencing both frameworks in spreadsheets, generating a 190-item obligation list with an estimated 60% overlap. A €35M fine ceiling made the stakes visceral.

How Beacon Helped

• 1Beacon's multi-regulation classifier automatically tagged the model against EU AI Act and GDPR simultaneously, surfacing 47 unique enforceable obligations (down from 190 raw items after deduplication).
• 2Clause-level change detection flagged the EU AI Act's August 2026 high-risk enforcement deadline 9 months in advance, auto-generating a remediation ticket in the team's ServiceNow IRM instance.
• 3The obligation engine cross-walked GDPR Art. 22 explainability requirements with EU AI Act Art. 13 transparency obligations — resolving both with a single documentation sprint rather than two separate projects.
• 4Predictive risk scoring identified that a planned model retraining on expanded demographic data would push the system into a higher-risk tier, allowing the team to restructure training data sourcing before the change shipped.

Outcomes

The Challenge

With 43 live AI models across credit underwriting, fraud detection, and algorithmic trading, the bank's Model Risk Management team faced simultaneous pressure from SR 11-7 (Fed/OCC model validation), DORA (digital operational resilience for their EU-facing trading desk), and a looming OCC examination. Each model had its own validation schedule in a separate spreadsheet; governance playbooks lived in a SharePoint folder last updated in 2021.

How Beacon Helped

• 1Beacon ingested the bank's model inventory and scored all 43 models against SR 11-7 Supervisory Guidance, DORA ICT risk requirements, and the bank's own internal policy — producing a tiered risk register in under 48 hours.
• 2The governance playbook generator produced model-specific validation protocols pre-mapped to SR 11-7 Section IV criteria, cutting the time for each Model Risk team validation report from 3 weeks to 4 days.
• 3Beacon's SAP GRC integration pushed obligation updates directly into the bank's existing GRC workflow — no duplicate data entry, no version drift.
• 4When the Fed published updated SR 11-7 supervisory expectations in Q1 2025, Beacon's clause-level change detection flagged the 11 affected models within 24 hours and drafted updated validation checklists automatically.

Outcomes

The Challenge

The company was preparing their first IND filing for a novel CD19-targeted CAR-T therapy in relapsed/refractory B-cell lymphoma. Their CMC team of four scientists had never submitted an IND before. FDA's 2024 draft guidance on CAR-T CMC had introduced new comparability study requirements for manufacturing process changes — requirements that their CRO's standard template did not reflect. With a $2.1M FDA filing fee at stake and an 18-month clinical timeline, a refuse-to-file (RTF) outcome was not an option.

How Beacon Helped

• 1Beacon's IND readiness scorer assessed the CMC package against current FDA guidance, flagging 12 gaps — including the missing comparability protocol and insufficient viral vector characterization data — before the package went to external regulatory counsel.
• 2The FDA corpus ingestion engine had indexed the 2024 CAR-T CMC draft guidance at clause level; it cross-referenced the startup's draft module against 67 specific guidance clauses and surfaced 8 that were unaddressed.
• 3Beacon's CMC draft generator produced a fully structured Module 3.2.S draft pre-populated with the company's existing analytical data, cutting initial document assembly from an estimated 6 weeks to 8 days.
• 4The SOT learning agent refined its recommendations based on two rounds of internal review feedback, improving suggestion relevance with each iteration.

Outcomes

The Challenge

The company was advancing three mRNA/LNP programs simultaneously: an oncology neoantigen vaccine (Phase I), an infectious disease prophylactic (pre-IND), and a rare disease enzyme replacement (IND enabling). Each program had a distinct regulatory pathway — FDA, EMA, and dual FDA/EMA for the rare disease program — with ICH Q6B, Q8, and Q11 guidelines all in play for LNP manufacturing characterization. The regulatory team of seven was spending 60% of their time on manual document harmonization across programs.

How Beacon Helped

• 1Beacon's multi-program architecture maintained separate IND readiness scores for all three programs, allowing the team to see cross-program regulatory debt at a glance and prioritize resources toward the programs closest to submission gates.
• 2The proteomics data module processed LC-MS/MS characterization data for the LNP formulations against ICH Q6B specifications, auto-flagging identity and purity attributes that required additional analytical method validation before FDA submission.
• 3For the dual-pathway rare disease program, Beacon mapped FDA 505(b)(2) and EMA Article 10 requirements in parallel, identifying 31 overlapping CMC obligations that could be addressed with a single harmonized data package.
• 4Beacon's 21 CFR Part 11-compliant audit trail covered all electronic records across all three programs — critical for an anticipated FDA pre-IND meeting where inspection readiness would be scrutinized.

Outcomes