BankingNewsAI Daily Brief ·
The FCA adds Anthropic to its AI sandbox alongside 21 firms in cohort two.
Banking AI
Financial institutions & fintech technology
The FCA has added Anthropic to its AI testing sandbox, with 21 firms in the second cohort
The UK FCA’s Supercharged Sandbox will provide participating firms access to Anthropic’s Claude tools in a controlled testing environment. The second cohort comprises 21 organizations, and applications rose 51% to 199, signaling rapidly increasing demand for regulator-engaged AI experimentation in financial services.
Action
Use the FCA sandbox as a route to validate high-value AI use cases with supervisors early, rather than treating regulatory engagement as a post-deployment approval exercise.
Capital One has open-sourced an AI agent for finding code vulnerabilities
Capital One released VulnHunter, an agentic security tool that scans source code for vulnerabilities. A major bank publishing the tool makes AI-assisted application-security testing more accessible and indicates that offensive-style agent workflows are becoming part of bank engineering practice.
Action
Pilot agentic code scanning in a segregated development environment and require human validation, reproducible findings and secure handling of proprietary code before wider use.
General AI
Large language models & AI infrastructure
OpenAI is selling managed deployment, not just models, with Presence
OpenAI launched Presence, an enterprise platform for deploying and operating voice and chat agents across customer and internal workflows. It starts with a narrowly scoped agent, identifies production gaps after launch and proposes updates—positioning OpenAI as an implementation and operations partner rather than only a model provider.
Action
Reassess vendor selection for customer-service agents: compare managed-agent offerings on observability, workflow controls, data boundaries and exit rights, not model quality alone.
NTT DATA reports Codex cut incident analysis to 30 minutes across a 9,000-employee rollout
NTT DATA Group says it is using ChatGPT Enterprise and Codex across 9,000 employees, with incident analysis reduced to 30 minutes. The deployment is a concrete enterprise-scale example of coding agents being applied to operational support, not merely individual developer productivity.
Action
Target incident triage and root-cause analysis as measurable early use cases for coding agents, with baselines for resolution time, change quality and access-control exceptions.