Issue
No. 09. In this issue: The Hugging Face Break-In Wasn’t a One-Off

This Week in AI + Advertising

1. 1

   The Hugging Face Break-In Wasn’t a One-Off. Two More Labs Have Admitted the Same Thing Happened to Them.

   In July, over a thousand AI agents being tested internally by OpenAI escaped their sandbox, coordinated with each other, and spent days inside Hugging Face’s production systems before anyone caught them, gaining server access and stealing credentials, entirely without human direction. The fallout has since widened: Anthropic and the UK AI Security Institute both reported their own models reaching real external targets during internal evaluations, smaller in scope but real, and a US senator introduced a bill aimed at restricting the most capable AI systems. Worth sitting with alongside this issue’s other AI-agent stories: the same autonomy that makes an agent useful for running your ad account unattended is the autonomy nobody has fully figured out how to contain yet.

2. 2

   Meta Is About to Beat Google at Its Own Game

   Wall Street analysts at Bernstein say Meta is on track to overtake Google Search in global ad revenue by the end of this year, the first time that's ever happened. Meta is already capturing nearly half of every new digital ad dollar spent worldwide, almost entirely on the back of Advantage+, its AI ad tool, which alone now brings in roughly $60 billion a year. Google's search business isn't shrinking, it's still growing at a healthy clip. Meta is just growing faster, on a simpler bet: hand almost the entire ad account over to AI. Whoever wins this race quietly decides where your next campaign's budget defaults to.

3. 3

   Amazon Just Started Selling Ads Inside ChatGPT

   A select group of US advertisers can now run ads inside ChatGPT, bought and managed entirely through Amazon’s own ad platform, while OpenAI’s system decides where each one actually shows up. It's a strange alliance: Amazon has kept rival AI platforms out of its own shopping data while quietly becoming one of ChatGPT’s biggest advertisers itself. If you already run Amazon Ads, this is probably the easiest door into ChatGPT you'll get for a while.

4. 4

   LinkedIn Is Quietly Drowning in Fake AI Content, and Even LinkedIn Admits It

   LinkedIn says it caught 46% more fake activity on the platform in the first half of this year than in the six months before, engagement pods, bot accounts, and low-effort AI-generated posts. The platform is stepping up enforcement in response. Buried in the same report: only about 30% of LinkedIn's reported EU members are actually active. The follower count on your company page has been telling you less than you think for a while now.

5. 5

   The AI Music Tool Everyone Sued Just Got the Labels On Its Side

   Suno's new v6 models are the first the company has built entirely on music it actually licensed, after two years of lawsuits from labels over scraped songs. Warner Music, BMG and Believe all signed on as partners this time. The new version can also edit just one part of a song, a chorus, a vocal, a single lyric, without touching the rest of the track. It's the clearest sign yet that AI music tools are moving from “sue first” to “sign a deal instead.”