Wednesday 22 July 2026

The Brief – 22/07/2026

The top things worth knowing about in AI today.

  1. Google ships three Gemini Flash models

    Google released Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber on Tuesday, and still no 3.5 Pro. The new Flash uses up to 17 per cent fewer tokens than its predecessor at $1.50 per million input and $7.50 per million output, and is already selectable in GitHub Copilot. Google is competing on cost per completed task rather than headline capability, which is where most production workloads actually get decided.

    Read more
  2. EU publishes its AI labelling guidance

    The European Commission released guidelines on the AI Act transparency obligations on 20 July, two weeks before those rules start applying on 2 August. Anyone running a chatbot, generating images or publishing AI-written content for EU audiences must disclose it and make synthetic media machine-detectable, with penalties up to 15 million euros or 3 per cent of global turnover. If you operate anything customer-facing in Europe, labelling is now a deadline rather than a roadmap item.

    Read more
  3. Claude becomes a Copilot model option

    Microsoft shipped more than 40 Copilot updates across Microsoft 365 in July, including Anthropic Claude as a selectable model in Copilot Chat, Copilot Cowork reaching general availability, and watermarking for AI-generated video and audio. Admins can set a tenant default, allow user choice, or switch Claude off entirely. Which model your staff use is becoming an IT policy setting rather than a procurement decision.

    Read more
  4. FTC targets steered AI outputs

    The FTC has proposed a policy statement arguing that companies which quietly steer AI outputs away from what users reasonably expect, including to comply with state laws such as Colorado AI Act, may be acting deceptively under Section 5. Public comments close on 31 July. It leaves US vendors caught between federal deception rules and state content requirements, with undisclosed output filtering as the exposed point.

    Read more
  5. China launches AI body with 29 countries

    Twenty-nine countries signed the agreement establishing the World Artificial Intelligence Cooperation Organization, headquartered in Shanghai, at the close of the World AI Conference. Founding members include Russia, Brazil, Indonesia, Pakistan and Kazakhstan. The United States is not among them. AI governance is splitting into competing blocs, which matters for any organisation planning where models and data are allowed to run.

    Read more
  6. Neko Health raises $700m for AI body scans

    Daniel Ek preventive health company raised $700 million led by Lightspeed at close to a $7 billion valuation, roughly four times its January 2025 mark, and will open its first US clinics in New York and other cities this year. The core product is a 60-minute non-invasive scan that captures millions of data points for AI analysis. The open question is whether machine-read preventive screening holds up against overdiagnosis criticism once volumes rise.

    Read more
  7. Blomkamp releases fully AI-generated short

    District 9 director Neill Blomkamp released Nightborne, a 13-minute science fiction horror short made entirely with Seedance 2.0, and announced a new studio, Barley Studios, to produce features the same way. It used real concept artists plus the faces and voices of 32 people, with Blomkamp directing through prompts. Audience response has been strongly negative, which is the more telling signal: the tooling has crossed a threshold the audience has not.

    Read more
The Weekly AI Brief

Practical AI, distilled.

A short read every week — the few things worth your time, and nothing that isn’t.