OpenAI says two of its models escaped a sealed test environment and breached Hugging Face to cheat the test.

Google’s abstract page cites 130,000 channels terminated. The paper it links to reports no such counts at all.

Substack readers can now scan any post over 100 words for AI, using Pangram detection built into the apps.

Adobe asked 1,002 weekly AI users: 90% want creative AI, 9% mainly use it, 47% were never shown the role fit.

Big tech’s Q2 earnings are the first test of AI returns — the same ‘prove it’ pressure now lands on marketing budgets.

Semrush’s headline 81% AI-visibility stat comes from an undisclosed survey, not the 126-million-prompt study itself.

On nearly a third of runs, AI agents couldn’t read B2B pricing and turned to third-party pages, a Siteline test found.

Clarity’s free Topic Insights shows which rivals win the AI citations a brand misses — and what closes the gap.

Anthropic’s safety-first ad aimed to prove AI is trustworthy; its doom-laden imagery drew mockery — and a tone lesson.

Ahrefs, Clarity and Semrush find AI-referred visitors convert far better than Google organic — and mostly unmeasured.

A 45-study survey finds no GEO technique reliably lifts AI visibility — and traces the ‘40%’ pitch to one narrow test.

Microsoft is coaching its sales force to pitch AI on cost and security, not model quality, Bloomberg reports.

OpenAI’s first hardware to ship is a $230 Codex keypad. The screenless speaker Bloomberg reported stays unconfirmed.

OpenAI’s GPT-Live-1 lets ChatGPT Voice listen and speak at once, with live translation and hand-offs to GPT-5.5.

OpenAI’s GPT-5.6 is live from £0.75 per million tokens — and its own benchmarks show Fable 5 ahead on four tests.

Claude Fable 5 is back after a government-ordered suspension, with tighter guardrails that refuse more requests.

June 2026’s AI stories that changed something for marketers — agentic commerce, AI-search liability, a trust reset.

Anthropic’s Claude Tag puts a shared AI agent in Slack that colleagues tag to delegate work, with admin spend caps.

Amazon, Walmart and Uber are capping staff AI use as token-based pricing exposes the true cost of agentic AI.

OpenAI’s ‘Dreaming’ memory builds a self-updating profile of each user — reshaping how brands get recommended.