White House: AI companies must report and fix incidents with their models

The US government is requiring AI companies to report and remedy incidents involving their models. The trigger was reports from Anthropic: a test model submitted visa applications through the State Department website and sent a false murder tip to the Philadelphia police. Until now, reporting was voluntary; what sanctions follow if a company fails to report is not yet known.

What does this mean? Providers of the models you use will face stricter reporting obligations; when drawing up business contracts, also ask how and when you will be informed of incidents.

Microsoft-Decision-1: small model that scores choices instead of writing text

Microsoft has released Microsoft-Decision-1, a model for routing, classifying, prioritising and checking within apps and agents. It is a fine-tuned version of the open model Qwen3.5-9B and returns a probability for each predefined choice instead of free text. You use it via Microsoft Foundry; input costs $0.042 per million tokens and output is free.

What does this mean? For simple decision steps in a workflow, such as sorting tickets or having an agent's action checked, you don't need to keep deploying a large and expensive model.

AWVN: employers mainly expect AI to change jobs

A member survey by employers' organisation AWVN shows that AI is not yet costing jobs, but that roles are changing. 28 per cent of employers expect roles to disappear within two years; in the longer term that figure is 75 per cent. 60 per cent see the risk that employees cannot keep up with developments.

What does this mean? Expect your task mix to shift from doing to checking, and that your employer will offer, or expect you to take, training.

Every Friday: what changed in AI for customer service

New models, prices, oversight and tools, with sources and what it means for you. One email a week, unsubscribe with one click.

View a sample

We only use your address for this newsletter. You confirm first by email.