Today's Feature

A surgeon’s cancer operation was cancelled last week after an insurer’s prior‑authorization workflow failed — and the surgeon filmed the aftermath. The short clip, shared by Dr Elisabeth Potter, has 681 likes and 219 reposts, crystallising a concrete cost: delayed care, angry patients, and immediate regulatory heat. For insurers the question is now fiscal, reputational and legal, not hypothetical. Two linked shifts explain why. First, carriers and vendors have spent five years chasing zero‑touch processing because straight‑through rates cut operating expense and win procurement pitches: CBS’s recent analysis of 1.3 billion claims flagged 19% in‑network denials in 2024, a reminder that throughput can mask outcomes. Second, agentic AI — systems that plan and execute multistep tasks — has moved from demo to production. Providers from AIG and McGill to niche startups are wiring agentic models into underwriting and capacity decisions (one specialty deal covers about $1.6 billion of premiums), and last week Clawchain.ai announced a developer platform to support autonomous agent infrastructure (launch: March 16, 2026). Agentic stacks can speed routine decisions; they also introduce new single‑point failures. Regulators have noticed. In 2025 HHS and CMS secured an industry pledge to speed and increase transparency in prior authorization processes. That commitment, made public to curb opaque delays, now frames a regulatory and political risk: a cancelled cancer op is not just an operational glitch, it is ammunition for reform. (cms.gov) The fraud backdrop amplifies the danger. INTERPOL’s 2026 assessment estimates global AI‑enhanced financial fraud produced roughly $442 billion in losses in 2025 and finds agentic techniques make scams far more efficient and profitable. Fraud is no longer a cottage industry: synthetic identities and coordinated scams are industrialised, often outpacing defenders. (interpol.int) SIU directors on industry panels report synthetic‑attempt volumes rising — one senior investigator put the uptick at roughly 30% year‑over‑year — and warn that the same agentic tools insurers deploy for triage can be turned against them. What should carriers do? Three compact, non‑bureaucratic moves. First, bake accountable human oversight into automation. Require hard “stop buttons,” auditable provenance for every automated decision and explicit contractual assignment of responsibility — including indemnities from vendors that sell “autopilot” promises. Procurement should reward provenance and liability‑sharing, not only latency statistics. Second, pair automation with aggressive fraud defence. SIUs need access to agentic tooling governed for explainability, plus consortium signals: anonymised feeds of synthetic‑identity indicators, payment‑rail anomalies and behavioural fingerprints. No single carrier will out‑innovate industrialised fraud alone; cross‑firm signal sharing with strong privacy controls is the defensive baseline. Third, market the fix. Sales and product teams should recast offers from “we speed approvals” to “we speed safe approvals — auditable, reversible, indemnified.” Brokers and employers weary of black boxes respond to clarity; vendors that map explainable decision paths will climb RFP shortlists. There is a commercial arithmetic to this prescription. Speed without safety invites punitive regulation and churn; safety with reasonable speed buys durable trust and pricing power. Banks long paid a premium for better controls; insurers that learn that lesson will capture higher margins. And a blunt truth remains: technology shifts advantage to whoever is prepared to absorb residual risk. If vendors refuse indemnity and carriers refuse accountability, patients and providers will pay the tab — and the next scandal will be larger. Optimise any two of speed, safety and accountability at your peril. The immediate task for industry leaders is practical: reintroduce human custody, harden fraud defences, and sell that restoration as the product. In an era of agentic AI and industrial‑scale synthetic fraud, trust — proved by audits and contracts, not slogans — will be the last defensible moat.