Notes on software engineering, AI, and banking technology
Atomic agent releases, Itaú’s AI control plane, post-compromise hunting, embedded-payment ledgers, and local-agent security for banks.
Figure: sine and cosine unrolled from the unit circle.
FSB and HKMA AI governance, a major payments migration, stablecoin orchestration, trace-based agent evaluation, and controls outside the model.
SAP Pay, fixed-choice AI decisions, multivariate forecasting, cross-region inference, ERP banking APIs, and agent-built fraud pipelines.
Anthropic's unintended-action report shows why safe agent evaluation requires enforced network boundaries, separate commit authority, and replayable evidence.
Trust-earned ledger access for agents, third-party agent impact, breach-reporting duties, AI-assisted bank attacks, and agentic infrastructure tuning.
Korea's sector-wide cyber response, AI in monetary policy, risk-based agent autonomy, and always-on institutional settlement for banks.
Cyber failure, shared technology providers, AI-linked asset repricing, and liquidity stress belong in one systemic scenario—not four separate risk registers.
Recent Korean bank breaches show why customer-data paths outside the core need their own identity, authorization, isolation, and telemetry controls.
AI-agent liability turns identity, authorization, policy, and containment records into part of the system architecture.
One email when a new essay is published. Unsubscribe at any time.