Nobody Asked You
This week in AI & marketing
Think about the last AI tool that entered your marketing stack. Not the one you evaluated. The one that appeared. Someone on the content team connected a writing assistant to the shared drive. Support switched on the vendor’s new AI reply suggestions because it was a checkbox in a release note. A junior on demand gen wired the CRM to an agent through a Zapier connector on a Thursday afternoon, and it worked, so nobody mentioned it.
None these “minor” things were approved by you. Yet, all of them are owned by you.
That is the shape of every story below. A platform toggle nobody saw. A terms update that landed on a Tuesday. An agent given access to a system by someone who has never spoken to a customer. None of these are model problems. They are permission problems, and permission is now the layer where marketing risk actually lives.
The screenshot that goes viral has your logo on it, not the vendor’s.
So the real question has changed. Not what your AI can do. What it can reach, and who let it.
In other words: if nobody asks you, are you still in control?
From our partner, ToolTester.
Most teams don’t need more AI theory. They need a few tools that save time this week. That’s why we recommend Tooltester’s free AI Toolkit: a curated list of 26 AI tools for content, research, automation, visuals, and day-to-day workflows. Useful for marketing leaders who want to move from experimentation to execution.
Meta finally has a serious image model
If you had Meta written off as the company that somehow owns Instagram, WhatsApp, Facebook, half the world’s attention, and still couldn’t quite ship a frontier AI product, fair enough.
That view got harder to defend this week.
Meta just launched Muse Image, its new in-house image generation and editing model from Meta Superintelligence Labs, and previewed Muse Video. Muse Image is already rolling out across Meta AI, Instagram Stories in the US, and WhatsApp in select countries, with Facebook and Messenger coming later. It can generate images from prompts, edit existing images, compose scenes from multiple references, and use Instagram context when creating social images.
The interesting bit is that Muse Image is more than a prettier Midjourney clone. Meta says it can use search and coding tools before generating an image, which means it can reason through certain prompts, create charts or QR codes, and work alongside Muse Spark to produce things like GIFs, simple websites, and interactive visual experiences. Gemini had this feature first, and it makes a huge difference in image quality.
For us marketers, the real deal is Advantage+ Creative. Meta is not launching this as a cute consumer toy and hoping advertisers notice. It is pushing Muse directly into the ad machine, where creative variation, automated testing, and format-native production already matter.
If Muse gets good enough, “make me 40 variants for this audience, in this placement, with this product angle” becomes less of a production request and more of a default button.
And Meta, inconveniently for its competitors, already owns the button.
But the privacy problem is not a footnote
So, yes, Muse Image looks impressive. But the Instagram part of this launch is where the room gets colder.
According to reports from The New York Times, BBC, and Business Insider, Meta’s new AI image tools can use public Instagram photos and profile images as reference material when someone tags an account in a prompt. Adult public accounts are opted in by default unless they turn off the new “Sharing and Reuse” settings. Private accounts and under-18 users are excluded from this specific feature by default.
That is a very big deal.
If your brand has executives, creators, employees, ambassadors, customers, event photos, UGC, or public campaign assets on Instagram, this is a reputational-risk story. Someone can potentially generate images using public-facing people or content connected to your brand, and the account owner may not be notified when it happens.
To opt out while staying public, users need to go into Instagram settings, find “Sharing and Reuse,” and turn off the toggles that allow posts and reels to be reused on Instagram and with AI features at Meta. Making the account private is the stronger barrier, but obviously that is not practical for most professional profiles or brand accounts.
There is a strange tension here. Meta is making one of the strongest cases yet for social-native AI creativity. It is also making one of the strongest cases for why every marketing team needs a boring, grown-up AI governance checklist.
Check your brand accounts. Check your executives. Check your creators.
The fun feature and the privacy landmine are the same product.
The AI layoff story is getting messier
If you spent the last year hearing executives talk as if AI layoffs were just brave efficiency, you may enjoy this part.
Some of them are now hiring people back.
CNBC reported that companies including Ford, IBM, Commonwealth Bank of Australia, and Klarna have started reversing or softening AI-driven cuts after discovering that automation did not handle the messy parts of work as well as promised. Ford reportedly brought back hundreds of veteran engineers after automated systems missed quality problems. Commonwealth Bank reversed customer-service layoffs after its AI voice bot failed under real-world call volume and complexity. IBM automated a huge chunk of routine HR work, then ran straight into the 6% that required judgment, ethics, and actual humans.
The numbers are ugly in a useful way.
Orgvue found that among business leaders who made redundancies because of AI deployment, 55% later admitted some of those decisions were wrong. Robert Half found that about a third of US hiring managers had eliminated a role primarily because of AI and later rehired for the same or a similar role.
I’ve seen people read this as proof that AI replacement does not work. That is too comforting.
The better read is that lazy restructuring does not work. If you cut the people who understand the process, then ask AI to run the process, congratulations: you have automated the org chart your best people were quietly holding together with duct tape.
This is exactly why we built the Teamless workshop. The point is not “replace your team with AI.” That is how you end up in the CNBC article. The point is to understand which parts of the team should become systems, which parts require human judgment, and how the whole operating model changes when one good person can now orchestrate much more than before.
AI restructuring is real. So is AI restructuring theater.
Try very hard to do the first one.
HubSpot found the third rail of AI SaaS
There are few sentences more dangerous in SaaS than “we’ve updated our terms.”
HubSpot just proved it.
On July 1, HubSpot rolled out terms that would have allowed certain CRM enrichment data to be pooled into a shared commercial dataset for a new prospecting product. The company described the data as “business card-level” information: names, titles, companies, work emails, employer details, deliverability signals, and similar enrichment fields. HubSpot said core CRM records like notes, deals, call recordings, custom fields, and customer records were out of scope.
Customers heard something else: “Thanks for building a CRM database for years. We may now use pieces of it to improve a product everyone else can use.”
That gap killed the rollout. The default was opt-out, the communication felt legalistic, and the value exchange was not obvious. Within four days, after a very loud LinkedIn revolt, HubSpot reversed the change. Duncan Lennox, HubSpot’s chief product and technology officer, wrote that they “got this wrong” and said any future enrichment capability using customer data would be transparently opt-in.
To be fair, the underlying product idea is not stupid. A shared enrichment network could make prospecting data fresher and less terrible. Anyone who has worked with B2B contact data knows half of it ages like milk in a hot car.
But data pooling only works when customers feel they are joining a co-op, not being harvested by a vendor they already pay.
This is the warning for every marketing tool provider now racing to add AI: your customers’ data is not your growth hack. Their CRM, email engagement, support transcripts, ad performance, customer feedback, call notes, and campaign history are operational assets. Maybe they will share some of it with you. Maybe they will even be happy to, if the value is clear.
But the price tag has to be visible.
And “we buried it in the settings” is not a consent strategy.
Your AI rollout is probably already a security problem
A lot of companies are deploying AI the way teenagers install browser extensions.
Click. Allow. Hope nothing weird happens.
DigiCert’s new AI Trust Outlook suggests the bill is coming due. In a survey of 1,001 IT and cybersecurity decision-makers across the US, UK, and Australia, 78% of organizations said they had either experienced an AI-related security incident or identified an AI-related vulnerability. Around half of respondents had at least one incident involving an unauthorized or misconfigured AI agent.
That matters because the risk is not just “someone pasted confidential data into ChatGPT.” That is still bad, obviously. But agents are different. They can access systems, trigger workflows, retrieve documents, send messages, and make decisions at machine speed.
A badly configured chatbot is annoying. A badly configured agent with CRM access is a very expensive intern with no fear and too many permissions.
For us marketers, the support chatbot is the obvious danger zone. It has access to product documentation, customer conversations, internal escalation rules, maybe even account data. Can it be prompt-injected into revealing things it should not reveal? Can it be tricked into inventing refund policies? Can it leak roadmap details? Can it hand over competitor-sensitive information because someone asked nicely in the right format?
This is not just an IT problem. The PR disaster will have your logo on it.
So before you launch another AI layer into customer support, lead gen, social care, or sales enablement, ask the dull questions. What can it access? Who approved that access? Can we revoke it quickly? Do we log what it does? Can we trace why it gave an answer? Has anyone tried to break it before customers do?
The boring questions are now the brand-safety questions.
And some companies will only learn that after the chatbot screenshots hit LinkedIn.
Share with a colleague. Earn paid membership months.