Liability risk from agents
Companies are waking up to liability as agentic AIs get access to corporate cards and internal databases — the conversation has shifted to treating coders as 'risk managers' rather than just builders. (x.com) Architectural mitigations like air‑gapped intent separation have been proposed to 'cage' agent execution and limit runaway access to sensitive systems. (x.com)
FIS announced an agentic‑commerce platform on January 12, 2026 in partnership with Mastercard and Visa to let banks identify and authorize AI‑initiated transactions. (fisglobal.com) FIS said the product will be available to issuers by the end of Q1 2026 and cited a McKinsey forecast that agentic commerce could generate up to $1 trillion in U.S. retail revenue by 2030. (fisglobal.com) Mastercard rolled out “Agent Pay” in April 2025 and described workflows where AI agents complete B2B purchases using tokenized virtual corporate cards. (mastercard.com) Legal advisers at DLA Piper flagged in a June 9, 2025 briefing that agentic systems act 24/7 at scale, creating novel disclosure, contract‑binding, and explainability obligations that demand new governance frameworks. (dlapiper.com) Willis Towers Watson’s February 27, 2026 discussion with Professor Anat Lior finds insurers are experimenting with AI‑specific coverage because traditional tech and cyber policies leave gaps for novel agentic harms. (wtwco.com) Reporting on workplace change shows developers are shifting toward orchestrator and manager roles—Business Insider noted this trend on March 18, 2026—and platforms like GitHub Agentic Workflows now include sandboxing, permission controls and explicit billing for agent runs. (businessinsider.com) Vendors are shipping air‑gapped agent platforms for regulated users—Arinox unveiled CommandCore for on‑prem, offline agent execution on February 17, 2026—yet academic research from September 2024 documents covert‑channel techniques that can still exfiltrate data from air‑gapped networks. (analyticsindiamag.com)
