Google withheld latest model from AISI
Testing, America first.
Happy Friday,
Today, confirmation that the era of US frontier labs giving the AI Security Institute pre-release access to test their latest models really is over. Google opened up its newest model to the US government, but not the UK in a break from the past. It follows a similar move from Anthropic and exposes the UK's lack of leverage.
The UK's AI ambitions mean little if the sectors that matter most can't take part. HMRC transaction data, NHS patient records, MoD intelligence systems — all sit in on-premises infrastructure, beyond the reach of frontier models. VAST DataEnclave changes that.
It brings leading AI models to sensitive data wherever it lives, using hardware that guarantees no party simply has to trust another. Every action is cryptographically verified and audit-trailed.
For sectors such as financial services, healthcare and government, that creates a path towards using more valuable data with AI while maintaining stronger control over security, sovereignty and intellectual property.
Happening Next Week 📆
To Birmingham: Conservative Party Conference begins in Birmingham on Sunday, but Monday is the main day for tech events. At 11:30am Re:State has a panel on "rethinking the regulatory state". Public First, Google and the Social Market Foundation discuss the AI skills gap at 1pm, while at 4:15pm Bright Blue hosts shadow BIST secretary Julia Lopez and Sovereign AI Unit chair James Wise to talk public service reform. From 7pm it's Startup Coalition's tech party.
At the seaside: Plaid Cymru's Conference starts today in the resort of Llandudno. We'll keep an ear out for any data centre related news.
On Wednesday it's the Society for Computers and Law Conference at DLA Piper in London, while on Thursday Air Street Capital is publishing its always excellent State of AI report.
In today's edition:
🚨 Google didn't give the AI Security Institute pre-release access to its latest frontier model, as the White House pushes America-first AI testing.