Self-serve ChatGPT ad buying opened across Europe on 31 August, with no published benchmarks for advertisers.

Google's AI Overviews now open in full on some desktop searches, and Search Console still shows no clicks.

Anthropic cut Claude Fable 5.1 cache reads 75% to $0.25 per million tokens, leaving list prices unchanged.

OpenAI’s GPT-6 Astra is built for CRM and admin work, but its own table shows 41.4% of workflows finished.

Slack Code puts AI agents in shared project channels the whole team can see, with marketing named as next.

Google's AI Mode switched to Gemini 3.8 Flash on 2 September, the third Flash model to run Search in six weeks.

LinkedIn says the 40% fall in slop views came from its classifiers, not the million member flags reported.

An Empirank audit found 11% of the businesses OpenAI recommended carried a claim the evidence contradicted.

Cached prompts make up most of the agent token burn, and they bill at about a tenth of standard input rates.

Seven of the FTC's last eight AI-washing cases targeted B2B claims, and the deck writers were charged too.

Publishers are pulling book deals over unverifiable authorship, leaving commissioned content without a standard.

Slack’s chief product officer says a private channel is a blind spot for every AI agent reading a workspace.

Comfort with autonomous AI stalled at 50% in StackAdapt’s survey, while 42% call in-platform advice generic.

ChatGPT now scopes one in six background searches to a single named site, and Reddit's citation share fell 86%.

The IAB’s new framework asks advertisers to label AI in ads only when it could mislead consumers, not every time.

Ahrefs has launched Letaido, an agent workspace on its own web index, led by a time-savings claim it hasn’t proven.

ChatGPT names brands in its own search query before it fetches a page, and cites them far more, one analysis finds.

Callosum’s $100m seed, the Sovereign AI Fund’s first equity deal, signals AI’s shift to cost-per-task economics.

Google recorded a 32% rise in pages hiding instructions for AI agents, and most published defences do not hold.

Anthropic has explained Claude's watermark. It marks only the words Claude chose, so light edits leave little.