OpenAI adds a mode for sensitive work

Lockdown Mode limits what ChatGPT can touch

OpenAI is adding a setting for people who do not want ChatGPT reaching too far.

Lockdown Mode disables live web browsing, web image retrieval, deep research, and agent mode. OpenAI says it is rolling out to self-serve ChatGPT Business accounts and some eligible personal accounts.

The important part is OpenAI admitting the tradeoff in plain terms. Prompt injections can still appear in cached web content or uploaded files, so this is risk reduction, not a clean fix.

I think that honesty matters because sensitive AI use is moving faster than most company policies. A lawyer, analyst, or security team working late under a desk lamp needs fewer surprise paths for data to leak.

The feature also shows where AI products are heading: more modes, more limits, more user choices about exposure. That makes the product less magical, which is probably healthy.

The open question is whether people will actually turn it on before something goes wrong.

Microsoft wants its own AI supply chain

Microsoft is putting more of Copilot’s future inside its own walls.

At Build 2026, the company announced seven MAI models, a new Cobalt 200 processor, agent tools across Windows, Azure and GitHub, and a quantum chip roadmap for 2029. Its flagship MAI-Thinking-1 model has about 35 billion active parameters and a 256,000-token context window, though it is still in private preview.

This matters because Microsoft has spent years selling AI through OpenAI’s models while carrying the risk of someone else’s roadmap. The message from San Francisco, under conference lights and keynote applause, was clear enough: Microsoft wants leverage.

I think buyers should read this as a cost and control move more than a pure model race. The strongest claim is Frontier Tuning, where Microsoft says task completion rose from 13% to 87%, but that still needs proof outside its own labs.

Copilot is not suddenly independent. It is getting an escape route.

Bogus claims now come with edited evidence

A cracked bumper photo now needs a second look.

Aviva says it found more than 18,400 suspect claims in 2025, worth £233m across its brands. Motor fraud made up more than seven in 10 bogus claims in its UK general insurance business.

The detail that matters is the shift from staged crashes to inflated repair bills, credit hire claims, injury claims, and AI-made evidence. That makes fraud harder to spot because it can sit inside a real accident, on a real form, under fluorescent office light.

Aviva is using AI and analytics with human oversight, which feels necessary given the tools now available to ordinary scammers. The risk is that honest customers face more friction because insurers have to treat more evidence as questionable.

The claim form is becoming a harder thing to trust.

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.