AI News

Week of 2 July 2026

Updated 2 July 2026

A busy week bookended by cheaper capability and firmer rules. Anthropic put near-top-tier performance into its default model at a fraction of the price, and Google opened multimodal video generation to enterprise developers, while the EU finished turning its AI Act simplification into law. Washington also reversed a fortnight-long export-control block that had taken Anthropic's most powerful models offline. Capability got cheaper and the rulebook got clearer in the same seven days.

The most striking signal was closer to home: a large UK study found workplace AI use has more than doubled in a year, to 73%, with the heaviest users pulling ahead on pay and promotion. For UK firms the message is that AI adoption has crossed into the mainstream, and the advantage now lies in using it deeply rather than merely having access.

The stories in detail

Models · Everyday agents

Anthropic's Claude Sonnet 5 brings near-flagship power to the default tier

Anthropic released Claude Sonnet 5 on 30 June and made it the default for Free and Pro users: a mid-size model with performance close to the larger Opus 4.8 at a fraction of the cost.
It is built to run agentically, planning and using tools like browsers and terminals autonomously at a level that recently needed bigger, pricier models, with introductory pricing of $2/$10 per million input/output tokens until 31 August.
For adopters it pushes capable, cheaper AI into the default tier, and Anthropic says it is also more resistant to prompt-injection and misuse than its predecessor, which matters if you are deploying agents on real workflows.

Anthropic - 30 June 2026 →

Platforms · Multimodal

Google opens multimodal video generation to enterprise developers

On 30 June Google brought Gemini Omni Flash, its natively multimodal model, to APIs in public preview, letting developers build custom video workflows from mixed image, audio, text and video inputs for the first time.
It arrived alongside a fast, low-cost image model (Nano Banana 2 Lite), part of Google's push to make multimodal generation cheap enough for production use.
For firms in marketing, training or internal comms, AI video generation is moving from novelty to an API you can build into a pipeline, worth assessing before commissioning the next batch of content the old way.

Google - 30 June 2026 →

Regulation · European Union

EU AI Act simplification becomes law

On 29 June the Council of the EU gave final approval to the Digital Omnibus simplification of the AI Act, completing the legislative process; it enters into force three days after imminent publication in the Official Journal.
It fixes the delayed high-risk deadlines (2 December 2027 for standalone systems, 2 August 2028 for embedded) and confirms bans on AI-generated sexual deepfakes and CSAM from December 2026.
For UK firms selling into the EU, the headline high-risk obligations now have firm, later dates, but watermarking of AI-generated content still lands this December, so that is the near-term item to plan for.

Council of the EU - 29 June 2026 →

Adoption · United Kingdom

UK workplace AI adoption has doubled in a year, to 73%

A large UK study by Public First with Google, published 30 June, found workplace AI adoption has more than doubled in a year, reaching 73% from 34% in 2025.
It reports a clear link between deep AI use and career progression: the heaviest AI users are markedly more likely to report strong performance reviews, promotions and pay rises.
The takeaway for firms: adoption is now mainstream, and a gap is opening between people and organisations using AI deeply and those merely dabbling, a reason to invest in real capability, not just access.

Google / Public First - 30 June 2026 →

Risk · AI as infrastructure

US reverses course: Anthropic's Fable 5 and Mythos 5 come back online

On 30 June the US Commerce Department withdrew the export-control order that had forced Anthropic to pull Claude Fable 5 and Mythos 5 for all foreign nationals; global access resumed from 1 July after a 19-day outage.
The original 12 June ban followed reports that a crafted prompt could bypass Fable 5's safeguards; its withdrawal ends a fortnight in which a frontier model was switched off worldwide by government order.
The episode is a live lesson in continuity risk: even top-tier AI can go dark overnight for reasons outside your control, so treat model portability and a fallback as part of any serious deployment plan.

CNBC - 30 June 2026 →

Past editions

Want this turned into action for your firm?

Book an introductory call