In today’s Digest, we discuss the UK safety body being denied access to Anthropic’s Mythos 5.1, the M+C Saatchi management buyout falling through, and Alibaba sending AI agents into rival apps.

UK safety body denied access to Anthropic’s Mythos 5.1

Anthropic has quietly drawn a line around its newest AI model, and Britain finds itself on the wrong side of it. Claude Mythos 5.1, described as the firm's most capable system for cybersecurity and life-sciences work, was launched without pre-release access for the UK's AI Security Institute. Vetted American organisations got in first. This is a first, Aisi, widely regarded as the world's foremost frontier-model tester, had reviewed every prior Anthropic release, including Mythos 5 back in April. 

Whitehall's unease is not just diplomatic bruising. Officials worry about what it means for national security and economic competitiveness if the UK is squeezed out of frontier AI access, and the Cabinet Office has ordered a rapid review. Some in government suspect the hand of the US administration is behind the decision, though that remains speculation rather than fact. Anthropic itself has declined to explain the move publicly, saying only that it is working with Washington to widen access at home and abroad.

M+C Saatchi MBO falls apart

A planned management buyout at M+C Saatchi's Australian and New Zealand operations has fallen through, ending months of quiet negotiation. The deal, backed by growth investment firm Parc, was expected to spin the local agency into an independent, founder-led business. Instead, the two sides walked away after due diligence turned up terms neither could accept. The agency's local office confirmed the collapse in a brief statement, framing the outcome as a mutual, if disappointing, conclusion rather than a breakdown. 

The buyout push itself had been led by a cross-section of the local leadership team; chief executive Dani Bassil, strategy chief Simon Wassef, creative head Jeremy Hogg, client chief Anita Zanesco, along with Remi Couzelas and Jack Playfair from the agency's Re and sport-and-entertainment divisions. Their bid to take the Australian business independent will, for now, stay on the shelf.

Alibaba sends AI agents into rival apps

Alibaba Cloud unit unveiled an upgraded version of QoderWake, a tool that spins up ready-made "digital employees" from little more than a one-line job description. What sets this update apart is where those AI workers can go: not just Alibaba's own DingTalk, but also ByteDance's Feishu and Tencent's WeCom. 

According to Alibaba, the digital staff slot in without friction, pulling context automatically from shared files, calendars and group chats.