HAPPY FRIDAY,
Mid-October, which means the days are getting shorter, the AI news cycle is getting longer, and this week's stories range from "mildly concerning" to "genuinely alarming". Let's get into it.
In This Week's AI Fridays:
- The first confirmed case of spyware delivered through targeted ads, and why every marketer running digital campaigns needs to read it
- Accenture capped Claude spending at $29/month. Some employees were spending $1,000.
- The ChatGPT traffic drop everyone panicked about was probably a tracking problem the whole time
- HubSpot is adding spend controls because consumption-based AI pricing is making budgets hard to manage
- Plus Google's Universal Cart, Cal AI's influencer playbook, and a handful of quick hits
Targeted Ads Delivered Zero-Click Exploits to Install Mercenary Malware
This appears to be the first confirmed case of mercenary spyware being delivered through the programmatic ad ecosystem using zero-click exploits, meaning targets did not need to interact with the ad at all. The ad tech infrastructure that marketers depend on to reach the right person at the right moment was used here to compromise devices.
The Highlights:
- Zero-click delivery means no user interaction was required; the exploit ran through ad rendering alone
- Mercenary spyware of this type typically costs millions to deploy, meaning the targeting was deliberate and the victims were specific
- The attack vector was standard programmatic advertising infrastructure, not a fringe or compromised channel
- This is described as a likely first confirmed case, which is the part that should give pause; confirmed does not mean first attempted
The Takeaway:
The ad tech supply chain has had transparency problems for years, and most of us in marketing have treated that as a brand safety and fraud issue. It is worth updating that mental model now, as the security stakes are higher than they were last week.
Accenture Tells Employees They Can Now Spend Only $29 on Claude Per Month
Accenture has capped employee spending on Claude at $29 per month. Some employees were previously running up to $1,000 a month in usage. this points to a governance/policy issue and Accenture is not the only firm where this tension is playing out.
The Highlights:
- The new cap is $29/month per employee, a significant reduction for the subset of users who were spending significantly more
- Some affected employees were using Claude for complex, multi-step work that is difficult to do at lower usage tiers
- The cap appears to apply broadly rather than tiering by role or use case
- This follows broader enterprise pressure to get AI tool costs under control as usage scales across large headcounts
The Takeaway:
When your heaviest users are the ones most disrupted by a spending cap, that is usually a signal that the policy was designed around average usage rather than value generated. The firms that figure out how to tier AI access by actual workflow impact are going to have an advantage over the ones that just set a flat number and call it a day.
AI in Marketing
ChatGPT Traffic Drop Appears To Be A Tracking Problem
The sharp decline in ChatGPT-attributed referral traffic that had a lot of marketers concerned turns out to be, in large part, a measurement artifact. ChatGPT links do not pass UTM parameters through in the way standard referrers do, which means the traffic was likely arriving without being properly attributed, showing up as direct or untracked rather than AI-referred. The volume was probably there. The visibility into it was not. For anyone making strategic decisions about AI-driven traffic based on what their analytics platform was showing, this is worth a revisit, as the picture in your dashboard is probably not the whole picture.
HubSpot Adds Spending Controls and Flexible Billing as AI Usage Grows
Consumption-based pricing made a lot of sense when AI usage was light and predictable. It makes considerably less sense when a handful of power users can increase a monthly bill by a surprising amount and nobody notices until the invoice arrives. HubSpot is rolling out spend limits and more flexible seat arrangements in response to exactly this kind of budget chaos, and it is a reasonable acknowledgement that the old model of "use more, pay more" needs guardrails to work inside a marketing org with a real budget. The interesting question is whether controls like these end up protecting budgets or just throttling the teams who were getting the most value out of the tools.
When It Came to Cal AI's Viral Success, a Small Army of Influencers Paid Off
Cal AI hit $30 million in revenue in two years before being acquired by MyFitnessPal, and the engine behind it was a deliberate, sustained influencer programme rather than a single viral moment. The strategy was to seed a large number of mid-tier creators rather than bet on a few big names, which kept costs manageable and kept the product in front of relevant audiences continuously. It is a useful case study, not because the approach is surprising, but because it is well-documented and it worked. For anyone building or marketing an AI product right now, the takeaway is less about influencers specifically and more about sustained, distributed visibility being worth more than one-time spikes.
Google's Universal Cart Is Live: How Carts Get to Your Store
Google's Universal Cart is rolling out across Search and the Gemini app, letting users add items from multiple retailers into a single cart and either check out inside Google or hand off to the retailer site. This is Google inserting itself into the purchase moment more directly than before, and it changes the e-commerce funnel for anyone running paid or organic shopping campaigns. The question is not whether your products show up in results; it is whether your checkout experience is optimised for users arriving mid-intent from an outside cart rather than from a traditional product page visit.
Quick Hits
- Anthropic is warning that small AI models could self-replicate across IoT devices, which is a sentence I would have assumed was science fiction about eighteen months ago. Read more
- Both Anthropic and OpenAI are reporting revenue figures that independent verification is having some trouble squaring; worth keeping in mind when either company tells you something is performing at scale. Read more
- HubSpot's data on AI Overviews suggests brands are either cited or invisible; there does not appear to be a middle ground, which makes the SEO strategy question somewhat less nuanced than people would prefer. Read more
- Benchmark is saying out loud that the VC norm against backing competing companies is breaking down under AI pressure, which tells you something about how fast the competitive field is shifting even by Silicon Valley standards. Read more
- Meta released Muse Spark Open this week, claiming it rivals or beats leading Chinese AI models on multi-step reasoning. Read more
- Technical detail on Muse Spark Open: 816 billion parameters, built for coding and extended agent tasks, and positioned as the most capable open-weight model currently available. That is a fairly large number of parameters. Read more
This article was obviously generated with AI but curated by a human, don't be weird about it.
Compiled by
Pete Bishop
Chief Innovation Officer, ZGM Modern Marketing Partners
Pete Bishop has spent the last two decades helping brands adopt new technology without losing the plot. He hosts Artificial Breakdown, a podcast and weekly newsletter that translates AI news into practical marketing decisions, and scans 50+ sources each week to write this issue.
Get this in your inbox
One email a week with the AI and marketing news worth your time.