AI Intelligence Trend Snapshot — 2026-09-30

Scan window: 2026-09-28T16:03:45Z to 2026-09-29T16:01:24Z.

No new trend met the evidence threshold. All seven active trends were reviewed. Trends #2, #3, and #7 gained supporting signals, but no status or confidence change is warranted.

Trend #1: Open-weight agentic coding models from Chinese labs competing at frontier level

  • Status / confidence: strengthening / Medium
  • Run review: No new independent signal was strong enough to change the assessment. Status and confidence remain unchanged.

Trend #2: MCP entering enterprise security & enforcement phase

  • Status / confidence: emerging / Medium
  • New evidence: MCP TypeScript SDK 2.2.0 requires machine-to-machine OAuth providers to declare an issuer and rejects authorization-server mismatches before token requests; list calls without a cursor now auto-paginate. Claude Code 2.1.284 also adds private_key_jwt IdP authentication, bulk MCP reconnect, and managed-model policy warnings.
  • Assessment: Enterprise identity, recovery, and default behavior keep becoming more explicit, but cross-vendor schemas and public production telemetry are still missing. Stays emerging / Medium.

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

  • Status / confidence: established / High
  • New evidence: Codex 0.159.0 adds in-flight steering and item-level history pagination. Qwen Code 0.24.7 moves workspace identity, execution receipts, and takeover/reboot recovery into stable.
  • Assessment: Runtime control planes and recovery paths continue to fill out, reinforcing an established trend. Cross-tool interoperability remains unresolved. Stays established / High.

Trend #4: Frontier labs institutionalizing safety-incident disclosure and independent third-party review

  • Status / confidence: strengthening / Medium
  • Run review: Anthropic still describes the METR incident review as planned, and no completed independent report was found. No new independent signal was strong enough to change the assessment.

Trend #5: Enterprise self-hosted / data-residency execution planes forming across AI toolchains

  • Status / confidence: candidate / Low
  • Run review: No new independent signal was strong enough to change the assessment. Status and confidence remain unchanged.

Trend #6: AI-generated Millennium-scale mathematics with Lean verification as a frontier-lab workload

  • Status / confidence: emerging / Medium
  • Run review: No new independent signal was strong enough to change the assessment. Status and confidence remain unchanged.

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

  • Status / confidence: strengthening / Medium
  • New evidence: Codex preserves filesystem denials on approved commands and protects .aws; Claude Code adds one-off permission for reads outside the workspace; Qwen Code adds execution receipts and post-reboot reconciliation.
  • Assessment: Controls keep moving into runtimes instead of depending only on model refusal. A unified policy schema and cross-runtime interoperability are still absent. Stays strengthening / Medium.

Invalidated / retired

No trend was newly invalidated or retired in this run.