Today's signal
Most infrastructure teams have already had an AI-agent incident, and the banks are now treating a shared platform and central guardrails as the way to scale AI safely.
Frontier AI
Databricks AI Index: companies are quickly moving away from relying on one AI model
Source Databricks
What happened
Data from more than 10,000 companies shows the share relying on a single AI model provider fell from about 64% in July to 28% by 2 November. Open-weight models, which companies can download and run themselves, grew from 7.3% to 30.9% of enterprise AI use in under four months. Databricks also found that no single model is best on both cost and quality, so companies need to test models on their own work.
So what
The question has moved from which vendor to pick to how quickly you can switch models safely. Because no single model wins on both cost and quality, enterprises need a way to test models on their own work and avoid being locked in.
In the enterprise
NAB gives its Ruby AI platform to more than 40,000 staff and sets up 50 AI teams, while ANZ sets up central control over its AI
Source iTnews
What happened
At its FY26 results, NAB described one AI and agent platform for the whole bank plus 50 teams led by the business. CEO Andrew Irvine said "we have to get it 100 percent right with AI." ANZ CEO Nuno Matos announced central oversight and control for the bank's growing use of AI agents.
So what
Two of the big four banks have landed on a similar model: one central platform with shared guardrails, and the business owning delivery. That is becoming the default operating model for AI at scale in large, regulated organisations.
Security & risk
Gartner: weak AI governance will cause major outages for 40% of infrastructure and operations teams by 2028
Source Gartner
What happened
Gartner surveyed 298 leaders who run infrastructure and operations in the US and UK. 95% had at least one problem caused by an AI agent in the past year, 57% say their controls over AI are weak, and only 15% have an automatic way to stop an agent that goes out of control. Gartner recommends seven fixes, including giving every agent its own tracked identity, building guardrails into the live system, setting spending caps for AI agents, and requiring human sign-off for high-impact actions.
So what
Weak governance is becoming an operational risk, not a paperwork issue. Building identity, guardrails, spending caps and human sign-off into the live system is more convincing than leaving those controls to a committee.