AI News

Week of 23 June 2026

Updated 23 June 2026

This week the theme was control: over the rules, over the models, and over who gets access. The European Parliament formally adopted its slimmed-down AI Act timetable, and the US government abruptly switched off Anthropic's most powerful models for all foreign nationals on national-security grounds. The rules and the geopolitics of AI tightened at the same time.

The build-out continued underneath. AWS pushed to make enterprise agents more governable and even gave websites a way to charge AI bots for content, while the talent war flared again as a co-inventor of the technology behind modern AI left Google for OpenAI. For UK firms the throughline is dependence, on foreign rules, foreign models and a fast-moving market, which makes portability, governance and a clear view of your own data and content more valuable than ever.

The stories in detail

Regulation · European Union

EU Parliament formally adopts the slimmed-down AI Act

On 16 June the European Parliament adopted the Digital Omnibus on AI by 423 votes to 57, locking in delayed deadlines for high-risk systems (December 2027 for standalone Annex III, August 2028 for embedded) and a ban on AI nudifier and CSAM-generating tools.
Final Council sign-off is expected before 2 August; until it publishes in the Official Journal, the original AI Act timeline stays legally in force, so those August dates remain real for now.
For UK firms in the EU market the practical read is more breathing room on high-risk obligations, but watermarking of AI-generated content still lands in December 2026, so plan for that regardless.

iubenda - 19 June 2026 →

Models · Geopolitical risk

US export controls force Anthropic to pull Fable 5 and Mythos 5

On 12 June the US government issued a national-security export-control directive barring all foreign nationals, inside or outside the US, from accessing Anthropic's most powerful models, Fable 5 and Mythos 5; Anthropic disabled both for everyone to comply, days after Fable 5's public launch.
The government cited a potential jailbreak that could expose the models' advanced cybersecurity abilities; Anthropic disputes that a narrow flaw warrants recalling a commercial model and says it is working to restore access. Other Claude models are unaffected.
The lesson for adopters is concentration and continuity risk: a single government directive can switch off a frontier model overnight, so build with model portability and a fallback in mind. For UK readers, note that all foreign nationals includes UK users.

Anthropic - 13 June 2026 →

Platforms · Enterprise tooling

AWS makes enterprise AI agents easier to govern

At its New York summit AWS expanded Bedrock AgentCore with features to connect agents to trusted knowledge, find and fix what goes wrong in production, and enforce controls that scale as agents grow more capable, plus a managed knowledge base for enterprise RAG.
It is part of a broad push to make agentic AI deployable with governance built in, the recurring theme as agents move from demos into production.
For firms weighing how to run agents safely, the big cloud platforms are increasingly competing on control and observability, not just raw capability.

Amazon Web Services - 17 June 2026 →

Industry · Talent

Transformer co-inventor Noam Shazeer leaves Google for OpenAI

Noam Shazeer, a co-author of the 2017 "Attention Is All You Need" paper that underpins modern AI and a Gemini co-lead, announced on 18 June he is leaving Google DeepMind for OpenAI.
It is a striking move in an intensifying talent war; Google had paid around $2.7 billion in 2024 to bring Shazeer and his team back from Character.AI.
For buyers it is a reminder that the frontier-model race is still fluid: today's capability leader can shift with the people, another argument against over-committing to a single provider.

CNBC - 18 June 2026 →

Content · IP economics

AWS lets websites charge AI bots for content

AWS added a capability to its WAF that lets content owners price, meter and charge AI bots and agents for access to their content and APIs, collecting payment at the network edge.
It is early infrastructure for a live question: how publishers and businesses get paid when AI agents, rather than people, consume their content.
For any firm whose value sits in proprietary content or data, how to license access to AI crawlers and agents is becoming a real commercial decision rather than a hypothetical.

Amazon Web Services - 17 June 2026 →

← Back to the latest AI News