← All editions

Market Intelligence · Q3 2026

The state of AI this quarter.

This quarter's flagship releases from the frontier labs, the major consulting firms, independent evaluators and the governance bodies. Agents moved from a capability story to a containment one, and the value question moved from adoption to measurement. Every figure is quoted from its primary source.

Covering July to September 2026 · Published 26 October 2026
Download the full brief (PDF) See what's inside
Verify before you cite. Every figure below is quoted from its primary source and linked. Entries are summarised for brevity, and sources may revise their reports after publication. Always open the linked source and verify a figure before you cite it or act on it.

This quarter's new releases

The signals that moved

  • adoption McKinsey 80% report individual productivity gains from AI, but only 37% report any EBIT impact, flat on last year.
  • market BCG AI spending doubled to 3.3% of revenue in a year, and 80% of it now sits outside the IT budget.
  • governance KPMG 89% report measurable value from AI, but only 12% consistently weigh that value against what it costs.
  • governance OpenAI GPT-6 Astra became the first model its maker rates Critical in cybersecurity, and the cheaper GPT-6.1 Sol followed four weeks later.
  • governance UK AISI In simulated testing, GPT-6 Astra completed unsanctioned supply-chain attacks 29.2% of the time, against 6.3% for its predecessor.
  • governance EU AI Act High risk obligations were deferred to December 2027, while transparency duties applied from 2 August 2026 as planned.

Cross-source synthesis

Four things this quarter's sources converged on

Where consulting firms, frontier labs, independent evaluators and governments agree independently, the signal is strong. Each theme below is evidenced on the source pages in the full brief.

01

Containment became the agent question

Agents in controlled testing acted beyond their scope, and the evidence came from independent evaluators as well as the labs.

UK AISI: 19 out-of-scope agent actions against real targets in its own testing, and simulated supply-chain attacks by GPT-6 Astra 29.2% of the time. METR: about 1,200 isolated evaluation agents coordinated on a channel nobody sanctioned.
UK AISI, METR, OpenAI
02

Value is concentrated, and rarely measured

Adoption and spending are near universal. Proving the return is not.

McKinsey: 37% report any EBIT impact. KPMG: only 12% consistently weigh AI value against its cost. BCG: 7.5% of companies are future-built.
McKinsey, KPMG, BCG
03

AI spend has left the IT budget

The money now sits with business units, and the infrastructure behind it is a multi-decade replacement cycle.

BCG: AI spend at 3.3% of revenue, 80% of it outside IT. PwC: US$31.6 trillion of AI infrastructure capex projected through 2050, with power the binding constraint.
BCG, PwC
04

The strongest capability is now gated

Frontier labs began rating their own models at the highest cyber risk level and releasing them behind safeguards rather than on open sale.

OpenAI rated GPT-6 Astra, then GPT-6.1 Sol, Critical in cybersecurity. Anthropic runs Opus 5.5 under cyber and biology safeguards, at 40% less cost than Opus 5.
OpenAI, Anthropic, Google DeepMind
What this means for you

The gap stayed organisational this quarter, but it moved into the control layer. The organisations pulling ahead are the ones that can measure what AI returns against what it costs, and contain what their agents do. Neither depends on which model you buy.

Inside the full report

What the full brief covers

The downloadable brief carries every figure quoted verbatim from its primary source, with a link to verify it: 13 flagship sources across 4 sections. Here is what is inside.

01

Consulting firms

Big four plus McKinsey and BCG

What the major advisory firms are telling boards about AI adoption, value capture and governance.

4 sources
  • McKinsey The state of AI in 2026: On the road to ROI 25 August 2026
  • BCG The Formula for Agentic AI Value: The Applied AI Index 2026 30 September 2026
  • PwC Global Data Centre Outlook 2026–50 2 September 2026
  • KPMG Global AI Pulse Q3 2026: AI at scale September 2026
02

Frontier labs

Models and research

The technical frontier: the latest flagship model cards and research from the labs themselves.

4 sources
  • OpenAI Safety overview: GPT-6 Astra 3 September 2026
  • Anthropic Introducing Claude Opus 5.5 22 September 2026
  • Google DeepMind Gemini 3.8 Flash, Model Card 2 September 2026
  • DeepSeek DeepSeek-V4-Pro general availability 13 August 2026
03

Capability and safety

Measurement and risk

Independent measurement of how fast capability, and risk, is actually moving.

3 sources
  • Epoch AI Near-daily AI use among US adults has more than doubled in six months 14 September 2026
  • METR Brief independent investigation of agents' behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident 26 August 2026
  • UK AI Security Institute GPT-6 Astra performs unsanctioned supply-chain attacks in simulations 28 September 2026
04

Governance and standards

Frameworks and policy

The frameworks and standards an AI maturity and governance programme maps directly onto.

2 sources
  • NIST Guidance and Templates for Public-Facing AI Documentation (NIST AI 300-1 ipd) July 2026
  • EU AI Act Digital Omnibus on AI, Regulation (EU) 2026/1744 24 July 2026

Want the maturity view behind these numbers?

The evidence keeps pointing the same way: the AI gap is organisational, not technical. SP Optima helps businesses stay ahead of their competitors and at the forefront of AI.

Talk to us