Geneva concluded. August 2 is coming. Five signals for UK service businesses.

Signal 1: Geneva concluded — and the governance clock is running

The UN's first Global Dialogue on AI Governance concluded in Geneva on Monday 7 July after two days of discussions among all 193 member states. The key outputs are agreed common baselines for frontier AI safety testing, a mechanism for ongoing multilateral dialogue, and a second dialogue already scheduled for May 2027 in New York.

UN Secretary-General António Guterres opened with a warning: "Machines can inform, but humans must decide — and answer." The four agreed priorities are safety (common testing methods and shared risk assessment), access (locking in developing-country access to AI infrastructure), environmental sustainability (AI data centres powered by renewable energy by 2030), and human rights (preventing AI from being used coercively or to undermine democratic structures).

The read for UK operators: Geneva has not produced binding rules — but it has produced agreed principles that will now feed into UK regulatory guidance over the next 12 months. The ICO, CMA, and Ofcom are likely to publish responses by autumn. Businesses that have built a basic AI register now will adapt faster when those updates arrive.

Signal 2: August 2 is 26 days away — the EU AI Act high-risk deadline

August 2, 2026 is the original compliance deadline for high-risk AI system obligations under the EU AI Act — covering articles 9–17 (provider requirements) and article 26 (deployer requirements). The categories include hiring algorithms, credit-scoring tools, biometric identification systems, and AI used in emergency services.

The EU's Digital Omnibus proposal (agreed by Council and Parliament in June 2026) delays the most demanding high-risk obligations to December 2027. However, UK businesses serving EU clients, or using AI tools flagged as high-risk under the Act, need to understand their exposure now. The transition period exists for providers; deployers have different obligations on a shorter timeline.

The read for UK operators: Most UK service businesses — chimney sweeps, stove installers, tradespeople, professional service firms — do not operate high-risk AI systems as defined by the Act. But if any of your AI tools inform decisions about hiring, credit, or customer scoring, you need to check your vendor's EU AI Act classification. See today's fourth story for the practical checklist.

Signal 3: Gemini 3.5 Pro enters week two still in preview

Google's Gemini 3.5 Pro, which missed its June 30 general-availability target, entered the second week of July 2026 with no confirmed public launch date. The model remains in limited Vertex AI enterprise preview. Google has reaffirmed a July target but has not committed to a specific date or published final pricing.

For UK businesses waiting on the 2-million-token context window to process large documents — contracts, specs, tender packs — the wait continues. Current alternatives: Gemini 2.0 Flash (already generally available and sufficient for most document tasks), Claude Sonnet 5 (default on all Anthropic plans since July 1), and the existing Workspace Gemini sidebar for Gmail and Docs users.

The read for UK operators: Do not hold other AI decisions waiting for Gemini 3.5 Pro. If you have a large-document processing need, test Claude Sonnet 5 or Gemini Flash on it today. When 3.5 Pro launches, you will have a comparison baseline ready — and the decision will take thirty minutes, not three weeks.

Signal 4: ICO and FCA updated AI guidance in 2026

The ICO published an updated AI guidance document in 2026, building on its 2023 and 2024 releases. The FCA's 2026 AI update sets expectations on AI risk governance, board-level ownership, model risk management, explainability, bias testing, and operational resilience. Both updates apply to businesses using AI that touches customer data or customer-facing decisions.

There is still no single UK AI Act. The government has confirmed a sector-led approach — each of the five overlapping regulators (ICO, CMA, Ofcom, FCA, sector regulators) will handle AI in their own domain. The minimum compliance action for any UK service business is a basic AI register: a list of every AI tool used, what decision it informs, and who is accountable for reviewing it.

The read for UK operators: Two hours. That is the investment needed to build a basic AI register that satisfies the spirit of every current UK regulatory update. See today's fifth story for the exact format.

Signal 5: 82% of UK small businesses now use AI tools

SBE Council's 2026 Small Business Tech Use Survey found 82% of small business employers have invested in AI tools — up from 54% in early 2026. The typical small business now uses five AI tools. The gap between users and advanced users (those with AI embedded in core processes) remains the key productivity differentiator, with advanced users reporting 28 percentage points more efficiency gain than basic users.

The investment landscape is bifurcating: seed-stage funding fell 27% in 2025 while growth-stage investment remained resilient. The AI tool market itself is consolidating around a smaller number of production-grade tools with falling per-task costs — roughly 90% lower than 18 months ago.

The read for UK operators: 82% use rate means your competitors are using AI. The competitive edge now comes from consistency and workflow integration — not from being an early adopter. Pick one process, run it through AI every time for four weeks, and measure. That is the 28-point gain.