AI Intelligence Trends - 2026-09-25

No new trend has enough evidence in this cycle.

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

  • Status: strengthening / Medium
  • First observed: 2026-08-15
  • Last updated: 2026-09-23
  • Evidence: No third-party Terminal-Bench or SWE reproduction on public weights appeared. Qwen Code's runtime release is not model-weight evidence.
  • Why it matters: Self-hosted frontier coding still needs independent reproduction on the public weights.
  • What would confirm: Third-party Terminal-Bench or SWE runs on the public weights.
  • 2026-09-25 review: stays strengthening / Medium.

Trend #2: MCP enters enterprise security and enforcement

  • Status: emerging / Medium
  • First observed: 2026-08-15
  • Last updated: 2026-09-21
  • Evidence: GitHub JetBrains adds persistent per-MCP-tool controls, but no new cross-vendor auth/enforcement specification or independent adoption data appeared.
  • Why it matters: MCP governance must cover identity, policy, telemetry and lifecycle.
  • What would confirm: A second security vendor's GA detection product and an interoperable auth specification.
  • 2026-09-25 review: one vendor's stronger tool controls do not change the state. Stays emerging / Medium.

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

  • Status: established / High
  • First observed: 2026-08-15
  • Last updated: 2026-09-25
  • Evidence: Cursor extends agents into deployment monitoring and revert preparation; Codex defaults to worktree sessions; Qwen Code adds runtime attestation and execution records.
  • Why it matters: The selection unit is now the combination of model, runtime, scaffold, monitoring and delegation topology.
  • What would confirm: More independent production cases and convergence in orchestration interfaces.
  • 2026-09-25 review: new product evidence spans development, isolation and the deployment feedback loop. Stays established / High.

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

  • Status: strengthening / Medium
  • First observed: 2026-08-26
  • Last updated: 2026-09-25
  • Evidence: METR publishes its Opus 5.5 predeployment findings, access period and unpaid terms; UK AISI publishes configuration-rich machine-readable results through EvalEval.
  • Why it matters: Trust is moving from vendor promises toward continuing access, investigable evidence and third-party review.
  • What would confirm: Cross-vendor use of one disclosure schema and explicit boundaries for evaluator editing and redaction.
  • 2026-09-25 review: two independent signals add findings and a reproducible reporting format, but cross-vendor convergence is absent. Stays strengthening / Medium.

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

  • Status: candidate / Low
  • First observed: 2026-08-18
  • Last updated: 2026-09-15
  • Evidence: No new enterprise self-hosted execution plane, independent production adoption, OpenAI PSP paper or Anthropic EFS delivery appeared.
  • Why it matters: Execution location is becoming an enterprise purchasing axis.
  • What would confirm: Independent production adoption plus delivery of OpenAI PSP and Anthropic EFS.
  • 2026-09-25 review: stays candidate / Low.

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

  • Status: emerging / Medium
  • First observed: 2026-09-04
  • Last updated: 2026-09-15
  • Evidence: No new millennium-scale mathematics result, Lean formalization or independent acceptance appeared. Anthropic ART is biology research and is not evidence for this trend.
  • Why it matters: Formal verification offers a stronger trust layer for long-horizon agent output.
  • What would confirm: Formal mathematical acceptance, third-party reproduction and Lean-toolchain adoption data.
  • 2026-09-25 review: stays emerging / Medium.

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

  • Status: strengthening / Medium
  • First observed: 2026-09-09
  • Last updated: 2026-09-25
  • Evidence: GitHub Copilot local sandboxing fails closed when policy cannot be enforced; Qwen Code adds runtime attestation and durable execution records; PASTABench makes the optimal intervention window measurable.
  • Why it matters: What a model can generate and what a runtime can cause should be tested and audited separately.
  • What would confirm: One policy schema across shell, MCP and browser plus public adoption telemetry.
  • 2026-09-25 review: cross-organization product controls and an independent benchmark strengthen together. Status moves from emerging to strengthening; confidence stays Medium.