The AI Brief for leadership communicators.
A weekly intelligence brief for speechwriters, executive advisors and leadership communicators, with my take on what each week actually means for the people who write and advise at the top.
The archive
12 issues publishedIl buono, il brutto, il cattivo.
A baroness speaks in the House of Lords in her own AI-recreated voice, an AI agent sends police a fabricated homicide tip, a UK legal services firm puts AI on its staff calls, Mumsnet's leaked prompt comes with an explanation that settles little — and a candidate answers AI videos that put words in her mouth.
Read the issueIl buono, il brutto, il cattivo
How leaders talk about AI may determine whether employees use it, NHS dementia tools head into real-world testing, agents fabricate evidence that they finished the job, Meta shares a home address it was never told to share — and why Intention, Context and Authority matter more than the perfect prompt.
Il buono, il brutto, il cattivo
AI boundaries, an ambitious attempt to extend cardiovascular care, fictional testimony in a criminal appeal, an 8,913-article fake newsroom, provenance for video — and the question of where human judgment still belongs.
Three Stories and The Truth
Full Fact catches three chatbots getting the news wrong, Pew finds the web increasingly contaminated by AI material, Imperial College gives the routine ECG a new trick, and Google Pinpoint gets a hands-on verdict.
Rules you can follow - July 14-27, 2026
The AP names AI tasks one by one, the EU's transparency obligations arrive 2 August, twenty-six Meta employees put AI telemetry in front of a federal judge, and the writing tic that gives all of us away turns out to be the small half of a bigger problem.
The privacy promise with an asterisk - July 6-13, 2026
Geneva hosts the UN's first government-level dialogue on AI governance, Meta opts millions into an AI feature and pulls it four days later, Anthropic hands its oversight body actual teeth — and Microsoft's enterprise privacy promise turns out to depend on which model happens to answer.
Adopted, then judged - June 29-July 6, 2026
The UN's first Global Dialogue on AI Governance opens in Geneva, a UN panel warns of widening inequality, Zuckerberg concedes AI agents are slower than hyped, consumers tire of AI content, and a chatbot reports a president died of rabies.
Who cleans up after the machines? - June 22-29, 2026
The EU's transparency rules land in August, Ford rehired 350 veteran engineers after over-trusting AI, a stray "Claude responded" surfaced in a congressional document, and AI agents become a brand-new audience.
The off-switch is the story - June 15-22, 2026
The model the U.S. banned was quietly un-banned a fortnight later, the G7 floats a fix to restore allied access, AI's "is any of this real?" problem deepens, and Norway draws a line for kids.
Fableus Interruptus - June 7-14, 2026
The most powerful model anyone could use was live on Monday and gone by Friday — plus the G7 at Évian, the Tokenpocalypse, and a real-time interpreter in your pocket.
What it means for us — week two
ChatGPT edges toward becoming a work surface, agents get identities and audit trails, Meta pushes business agents into messaging, and NotebookLM earns a closer look.
What it means for us — week one
Altman keeps his own voice, the "AI penalty" goes peer-reviewed, the EU comes for synthetic content, and Copilot agents start talking to each other.
No issues match that search yet — try a broader term.