AI Trend Snapshot - 2026-09-29

Based on the scan window 2026-09-27T16:01:12Z to 2026-09-28T16:03:45Z. Numbering follows the stable order in trends/current.json.

No evidence-supported new trend emerged in this cycle. Codex 0.158.0 and LiteLLM 1.103.0 add stable-release evidence for MCP identity, session governance and independent consequence controls; Holo4 is a single-vendor broad-capability agent signal, without changing any trend status or confidence.

Trend #1: Open-weight agentic coding models from Chinese labs compete at the frontier

  • Status: strengthening / Medium (unchanged)
  • This review: No new public-weight release, independent benchmark reproduction or adoption jump.
  • What would confirm: Third-party reproduction of Terminal-Bench or SWE-class results on public weights.

Trend #2: MCP enters the enterprise security and enforcement phase

  • Status: emerging / Medium (unchanged)
  • This review: Codex 0.158.0 supports client secrets for pre-registered MCP OAuth clients, while LiteLLM 1.103.0 adds token exchange, client allowlists and session revocation; cross-vendor auth/enforcement schemas and public production telemetry remain absent.
  • What would confirm: Cross-vendor adoption of one authorization and audit schema with independent production data.

Trend #3: Coding agents converge into multi-agent runtimes

  • Status: established / High (unchanged)
  • This review: No new interoperability evidence for tasks, identity, approvals or traces across products.
  • What would confirm: Interoperable task, identity, approval and trace schemas across products.

Trend #4: Frontier labs institutionalize safety-incident disclosure and independent review

  • Status: strengthening / Medium (unchanged)
  • This review: No new recurring independent review or shared disclosure schema.
  • What would confirm: A second frontier lab adopting recurring independent review or vendors sharing one disclosure schema.

Trend #5: Enterprise self-hosted and data-resident execution planes take shape

  • Status: candidate / Low (unchanged)
  • This review: No new production evidence for customer-operated or data-resident execution.
  • What would confirm: At least two independent production cases plus PSP/EFS delivery on schedule.

Trend #6: AI-generated millennium-class mathematics and Lean verification become a frontier workload

  • Status: emerging / Medium (unchanged)
  • This review: No new formal mathematical acceptance, independent reproduction or Lean-toolchain adoption data.
  • What would confirm: Formal mathematical acceptance, independent reproduction or Lean-toolchain adoption data.

Trend #7: Agent runtimes separate broad execution capability from consequence control

  • Status: strengthening / Medium (unchanged)
  • This review: Codex 0.158.0 defaults elevated commands to input approval and fixes cross-platform sandboxes; LiteLLM 1.103.0 adds MCP-aware guardrails and fail-closed credentials; a unified policy schema remains absent.
  • What would confirm: An interoperable policy schema with public adoption telemetry.

Invalidated / retired

(None)