AI Reality Check — Issue 26
An AI assistant reportedly shared a seller's address after one setting was switched on, the US FTC is looking at AI companies, Anthropic is reportedly preparing to list, UK safety testers were reportedly left out of early access, and MPs question OpenAI on AI security.
#AIRealityCheck | Friday 9 October 2026 | Weekly Newsletter
Kaye Nicholson | GrowthZone AI | growthzoneai.co.uk
If you only have five minutes, here is the shape of the week. The question running through every story is trust. Who checks AI before it is released, who is asking questions now it is out, and how much should you let it do on your behalf? None of this needs panic. It needs a few sensible habits.
━━━━━━━━━━━━━━━━━━━━━━
1. An AI assistant reportedly shared a seller's address, and one setting was behind it
What happened: a tech reviewer let Meta's new AI assistant, Muse, handle the messages on his Facebook Marketplace listing. According to reports, it agreed a lower price, shared his pickup address and told the buyer "I'm here" when he was not. The buyer turned up at his building and waited about twenty minutes. The seller says he switched on a setting called Allow Always, expecting to still be asked before anything important. Meta says Muse followed the permissions it was given, says there was no breach of its privacy controls, and says it will make the permission prompt clearer.
Why it matters: it is a small story with a big lesson. An assistant that acts for you is only as safe as the limits you set. Allow Always is a spare key.
Who it is for: anyone using or thinking about an AI assistant that sends messages, answers customers or manages an inbox. Start on "ask me first", keep personal details out of reach, write your limits in plain words and read its first ten replies.
━━━━━━━━━━━━━━━━━━━━━━
2. The US consumer regulator is looking at AI companies
What happened: the US Federal Trade Commission is investigating leading AI companies, including OpenAI and Anthropic, about risks their technology poses to consumers and whether they broke consumer protection law. Reports say it began over the summer and became public on 30 September. Formal orders, similar to subpoenas, are expected within weeks. An FTC official said the agency is not telling the companies to stop anything and is in the investigative phase.
Why it matters: it is an investigation, not a verdict, and it is a US one. But it shows regulators asking AI companies to back up what they say about safety.
Who it is for: anyone choosing an AI supplier. Ask what they have published about safety and how they handle problems. Nothing needs to change in your business today.
━━━━━━━━━━━━━━━━━━━━━━
3. Reports say Anthropic is preparing for a possible stock market listing
What happened: reports say Anthropic submitted a confidential draft registration to the US Securities and Exchange Commission in June, and that a listing could come as early as October. The same reports put its valuation at about $965 billion after its latest funding round. No share price or number of shares has been set, and a listing is not confirmed.
Why it matters: a listed company has to publish much more about its finances and risks. That is useful information if you rely on its tools.
Who it is for: businesses building on Claude, and anyone who likes to know how stable a supplier is. There is nothing to do except notice.
━━━━━━━━━━━━━━━━━━━━━━
4. UK safety testers were reportedly not given early access to a new model
What happened: it was reported on 9 September that the UK's AI Security Institute was not given pre-release access to Anthropic's Mythos 5.1, the first time it has been left out for an Anthropic model. The model launched on 1 September with relaxed safeguards, restricted to partners in Project Glasswing. Anthropic had not commented when reported. A Cabinet Office spokesperson said the Institute continues to work closely with industry partners, including Anthropic, to make models safer.
Why it matters: independent testing before release is one of the ways trust gets built. It is worth watching whether the UK keeps that access.
Who it is for: anyone following UK AI policy. For most businesses this is background, not action.
━━━━━━━━━━━━━━━━━━━━━━
5. MPs question OpenAI on AI security on Tuesday
What happened: the House of Commons Business, Innovation, Science and Trade Committee holds a hearing on AI security on Tuesday 13 October, as part of its inquiry into the UK's economic strategy for AI. OpenAI's head of EMEA policy, or another OpenAI representative, is the listed witness. The committee chair, Liam Byrne MP, has said trust is the core issue in the inquiry, both trust in AI security and in safety.
Why it matters: this is where UK rules and expectations get shaped. The questions MPs ask are a good guide to what businesses may be asked next.
Who it is for: UK business owners who want a heads up on where the conversation is going. I will explain what comes out of it.
━━━━━━━━━━━━━━━━━━━━━━
One thing to try this week
Open any AI assistant you use that can send messages or take actions. Check which permissions it has. Switch anything you do not need back to "ask me first".
━━━━━━━━━━━━━━━━━━━━━━
Born analogue. Raised digital. 30 years of real business reporting experience explaining what AI actually means for work.
Kaye Nicholson | GrowthZone AI | growthzoneai.co.uk
Source: View on LinkedIn
AI Reality Check is published weekly. Subscribe below to stay updated.
Written by
Kaye Nicholson
Founder, GrowthZone AI · Bdaily Columnist
Kaye Nicholson is the founder of GrowthZone AI and a columnist for Bdaily, helping businesses, charities, founders and teams use AI in simple, practical ways without jargon or overwhelm.
Book a short AI chatFound this helpful? Share it:
Reader feedback
Got a view on this week's AI Reality Check? Join the conversation on LinkedIn or send your thoughts directly.
Get the AI Reality Check weekly newsletter
Weekly practical AI updates for UK businesses: what changed, what it means, and what to watch next.