AI in June 2026: a frontier model switched off and Microsoft's push to govern agents
June showed both sides of the agent era. Microsoft set out how businesses should govern fleets of AI agents, while a US government order took two newly launched frontier models offline within days of release. Meanwhile, Europe finalised its revised AI Act timeline and the UK announced billions in AI investment. Here is what happened, and what it means for your organisation.
The US government orders Anthropic's newest models offline
On 9 June, Anthropic launched Claude Fable 5 for general use and opened restricted access to the more powerful Mythos 5 through Project Glasswing, its cybersecurity programme for critical infrastructure partners. Days earlier, on 2 June, Anthropic had expanded Glasswing access to 150 additional organisations.
On 12 June, US Commerce Secretary Howard Lutnick issued an export control directive over security concerns about the models, and Anthropic disabled both of them. On 26 June, according to Axios, the Commerce Department said Anthropic had made "significant progress" on the risks identified and allowed Mythos 5 access to resume for a government-approved list of organisations. Fable 5 remained unavailable at the end of the month.
What it means for your business: This is reported to be the first time a government has forced a publicly deployed frontier model offline, and it happened within days of launch. Any workflow that depends on a single model can be disrupted by events outside your control, whether regulatory, commercial or technical. Design AI systems so that you can switch to an alternative model with minimal rework, and keep a tested fallback for anything business-critical.
Microsoft builds the control layer for agents
At its Build conference in San Francisco on 2 and 3 June, Microsoft focused heavily on running and governing AI agents at scale. Announcements included the general availability of the Agent 365 SDK, which builds observability, access controls and compliance enforcement into custom agents, and a preview of the Agent 365 Agent Registry, which surfaces unmanaged agents detected across Microsoft Defender, Entra and Intune, including coding assistants and MCP servers.
Microsoft also made Windows 365 for Agents generally available, giving agents isolated cloud PCs governed by Intune policies, and previewed Purview controls that detect and block sensitive data in prompts before it reaches an AI model. Its Foundry platform was positioned as the place to deploy agents into production.
What it means for your business: Many organisations already have more agents running than they realise, from coding assistants to tools staff have connected themselves. If you are a Microsoft 365 business, these features offer a practical route to discovering what is in use and putting sensible limits around it. A simple agent inventory, with an owner and a list of permissions for each, is a good place to start.
The EU signs off its revised AI Act timeline
The changes to the EU AI Act agreed in May completed their passage in June. The European Parliament voted to adopt the Digital Omnibus on AI on 16 June, and on 29 June the Council of the EU gave its final green light. The package moves obligations for stand-alone high-risk AI systems to 2 December 2027 and for AI in regulated products to 2 August 2028. It also adds a ban on AI systems that generate child sexual abuse material or non-consensual intimate imagery, and extends some small-business relief to small mid-cap firms. Publication in the EU's Official Journal was expected in July, with the changes taking effect shortly afterwards.
What it means for your business: The revised dates are now effectively settled, so you can plan with confidence. Other parts of the Act, including the rules on prohibited practices and general-purpose AI, already apply and were not delayed. If you sell into the EU, now is a good time to turn your AI inventory into a clear compliance plan with owners and dates.
The UK announces more than £6 billion of AI investment
As London Tech Week closed on 12 June, the UK government announced more than £6 billion of new investment and around 8,000 jobs. Highlights included a £1.1 billion AI Hardware Plan to back chip firms and computing power, AMD committing £2 billion over five years, Nebius investing £1.7 billion in UK infrastructure and Amazon investing over £1 billion with up to 4,000 jobs. The government also announced new AI skills support for young people and the first AI Adoption Summit, bringing together government, technology companies and unions.
What it means for your business: More UK-based computing capacity should make it easier to keep data and AI workloads in the country, which matters for regulated sectors and public sector suppliers. The focus on adoption and skills also signals that government support for businesses putting AI to work is likely to keep growing, so it is worth watching for programmes you can tap into.
Practical takeaways
Make sure every business-critical AI workflow can switch to an alternative model, and test that fallback.
Build an inventory of the AI agents in use across your organisation, with an owner and permissions for each.
Use data loss prevention controls to stop sensitive information reaching AI tools by mistake.
If you trade with the EU, turn the confirmed AI Act dates into a compliance plan with clear owners.
Keep an eye on UK adoption and skills programmes that could support your AI plans.
If June's events have you thinking about resilience, governance or where to take your AI plans next, we would be glad to help you work through it. Book a call with the Manta Global AI team and we can explore what is possible for your business.
Comments