What it means for us

Altman tested AI replies — then went back to writing his own

The CEO of OpenAI personally tested sending replies labelled "from Sam's AI," decided it felt wrong, and went back to answering messages himself. The man who built the thing that supposedly makes human communication optional decided his own voice was worth keeping.

Use this the next time a client asks why authentic executive voice still matters — or the next time they suggest you just let the machine handle it.

The "AI penalty" is now peer-reviewed

We've all felt it. Now there's a study. Communication perceived as AI-generated is rated as less trustworthy, less authentic, and less useful for knowledge transfer. The researchers are calling it the "AI penalty."

You can stop saying "I just have a feeling about this" and start citing a source.

"AI won't fix your leadership communication — but it might expose it"

Gallup data shows 95% of employees trust leaders who communicate clearly and decisively. This piece argues AI doesn't fill the gap — it illuminates it.

Useful reading before your next conversation with a leader who thinks a better prompt will solve a credibility problem that is actually a character problem.

EU AI Act synthetic-content labelling: what enterprises need to know now

The EU wants a label on every piece of synthetic content — clear, repeated, and persistent. If your organization produces AI-assisted executive statements, speeches, or press materials for European audiences, your legal and comms teams need to be in the same room about this. Soon. Companies such as TrueScreen (illustrative only) are already gearing up.

Deepfakes have entered routine business processes

Synthetic media is now showing up in payment approvals, support requests, and executive communications — and the "familiar voice test" is no longer reliable, because voice cloning has quietly gotten very good.

If your organization's crisis-communications plan doesn't address this yet, that's a board-level gap dressed up as an IT problem.

CEOs aren't worried about AI risk — which is itself a risk

BCG finds most CEOs are personally unbothered by AI risk. The authors call it a governance blind spot (uhh, yeah). The practical problem for those of us advising leaders is simpler: you cannot help someone communicate credibly about something they haven't genuinely reckoned with. Note: story is from March, 2026 but included as it's relevant today.

Time to get your leader to take the topic seriously — before they're forced to.

Copilot Studio agent-to-agent communication goes generally available

AI agents inside M365 can now pass tasks to each other automatically — drafting, routing, flagging, approving — without a human touching the file at each step. This is not a future scenario. It is rolling into enterprise environments now. The question of where human judgment must remain in the loop is no longer theoretical. It is this week's problem.

Copilot Notebooks: mind maps and a unified workspace

Copilot Notebooks got a meaningful update this week — mind mapping, smarter inputs, a unified workspace. It's a reasonable moment to take another look. For research, message architecture and briefing prep inside a managed environment, it could be a tool worth a ponder.

NotebookLM: three incoming features worth watching

Google is rolling out Personal Preferences, external Connectors, and Canvas — which turns source documents into timelines, explainers and visual outputs. Canvas in particular shifts NotebookLM from a reading tool to a drafting workspace. If you use it for speech research, this is the update you've been waiting for.

Gemini 3.5 Flash repriced — the era of "basically free" AI is closing

Google's Gemini 3.5 Flash just got significantly more expensive. If you or your clients have built comms workflows on the assumption that AI processing would stay cheap, it's time to revisit that math. It's not just Google, folks - this is a common message from across the industry that the infrastructure pricing holiday is ending.