Weekly signal

This briefing covers legal and regulatory signals about AI agents ("agentic AI") from 2026-10-12 through 2026-10-20. Three material items moved the regulatory needle: an international supervisory menu for banks (FSB final report, 15 Oct 2026), China’s draft export-control revision that explicitly names "AI agents" and requires pre-export registration (MOFCOM/MOST Announcement No. 48, 16 Oct 2026), and U.S. enforcement escalation where the FTC served civil investigative demands on OpenAI and Anthropic (mid‑Oct 2026). These events shift the mix from guidance to near-term operational obligations and enforcement risk for agent deployments.

What changed

1. Financial Stability Board (FSB) published a final "Sound Practices for Responsible Adoption of Artificial Intelligence" on 15 October 2026. The report keeps 12 non‑binding practices and adds operational guidance for highly autonomous/agentic systems (defining expectations for permissions, human‑oversight models and stop/kill controls). It is intended as a supervisory benchmark for financial institutions across G20 jurisdictions.

2. China’s Ministry of Commerce and Ministry of Science and Technology issued Announcement No. 48 on 16 October 2026: a draft revision to the "Catalogue of Technologies Prohibited or Restricted from Export" that adds 40 tech items (23 in AI) including large models, compute clusters and explicitly "intelligent agents." The notice opens public comment from October 16 to November 15, 2026 and imposes a temporary pre‑export registration requirement effective on publication. Open‑sourced/public domain technologies are carved out.

3. The U.S. Federal Trade Commission (FTC) escalated from public inquiries to formal evidence‑gathering: the agency served civil investigative demands (CIDs)/subpoenas to OpenAI and Anthropic in mid‑October (reported 14–15 October 2026). The demands seek documents on safety incidents, safeguards, marketing and privacy—and signal possible future enforcement under consumer‑protection laws. CBS News and Axios confirmed the agency’s action.

What to do with it

• Financial institutions and regulated firms: map agentic use cases to the FSB’s 12 practices immediately. Treat the FSB report (15 Oct 2026) as a supervisory checklist: inventory agentic systems, assign accountable owners, document permissions and human‑oversight modes, and implement interrupt/kill controls and traceability. Start board‑level briefings and evidence trails now.

• Companies that develop or transfer AI tech to/from China: treat Announcement No. 48 (16 Oct 2026) as effective for registration. For any planned transfers of model weights, source code, or agent tech, register via MOFCOM’s platform before export and use the Oct 16–Nov 15 comment window to seek clarity on thresholds. Consider splitting product delivery: API access vs export of weights/source code has different regulatory exposure.

• AI product teams and U.S.‑facing vendors: expect enforcement scrutiny. Preserve incident logs, internal testing records, change histories and marketing materials; be ready to respond to CIDs and to tighten disclosures and user controls. If you operate agentic features, document human oversight modalities and emergency stop mechanisms.

(See sources below for the primary documents and initial reporting.)