BankingNewsAI Daily Brief ·
Fannie Mae’s AI-governance deadline exposes critical control gaps across the mortgage industry.
Banking AI
Financial institutions & fintech technology
Fannie Mae’s AI-governance deadline is exposing mortgage-industry control gaps
A Fannie Mae AI-governance deadline has arrived, with mortgage firms split between organizations with mature AI inventories and controls and those still identifying where AI is used. The development turns AI governance from a policy exercise into an operational eligibility and counterparty-management issue for mortgage lenders and servicers.
Action
Require a mortgage-business attestation that every AI use case touching Fannie-related workflows has an owner, documented purpose, testing evidence, human escalation path and vendor controls.
General AI
Large language models & AI infrastructure
Meta’s Muse Spark 1.2 reached frontier-tier finance-agent performance at a fraction of leading-model cost
Meta’s Muse Spark 1.2 entered Vals Index’s top five at $0.69 per test and became the first model reported above 60% on Finance Agent v2, at $0.77 per test. Vals compared that with prior leader Opus 5 at $5.12 per test, while reporting roughly twice the speed. This makes cost-adjusted model routing—not simply selecting the highest-ranked general model—material to AI vendor and platform decisions.
Action
Benchmark Muse Spark 1.2 against your current fraud, operations, research, and internal-service workflows, with cost per completed task as a required procurement metric.
Amazon Bedrock adds policy controls that can enforce agent sequencing, exposure caps and approvals
AWS added temporal policies and rate-limiting controls to Bedrock AgentCore. The controls can require actions to occur in a specified order, cap financial exposure, prevent unsupported workflow steps and route high-value actions to human approval rather than relying only on prompt instructions.
Action
Have engineering evaluate whether policy-at-runtime controls can enforce segregation of duties, transaction thresholds and approval gates in planned AWS-hosted agent workflows.
The Ninth Circuit narrowed the CFAA theory used to block Perplexity’s commerce agent
The Ninth Circuit vacated a preliminary injunction that had barred Perplexity’s Comet browser and related agentic-commerce activity under the Computer Fraud and Abuse Act. The ruling is an early legal marker for agents acting on websites and may narrow one route businesses use to challenge automated AI interactions with their services.
Action
Review digital-channel terms, authentication design and bot defenses rather than assuming CFAA claims alone will deter third-party agents from accessing customer-facing services.