AI Leadership Weekly

Top Stories

OpenAI readies tech for native checkout tracking
OpenAI is preparing native checkout and order tracking inside ChatGPT, which points to a shift from chat toy to transactional platform. App build sleuths spotted a new Orders setting and hints that users will be able to add cards and track purchases right in the app. Voice Mode is also being tweaked, now launching in the main window, while long-teased screen sharing seems to have stalled.

• Native checkout and tracking. OpenAI is testing an Orders section that could "let users add credit cards directly to ChatGPT" and handle "order tracking" in-app. It will likely debut in the US first, according to the report.

• Voice mode adjustments. Voice Mode now opens within the main app window, which is a small but telling UX change. The previously demoed screen-sharing feature appears to be "sidelined", two years after its splashy demo.

• Parental controls incoming. Work is underway to "bring parental controls to the ChatGPT web interface", though timelines are unclear. This tracks with industry pressure on safety in education and youth use.

Why it matters: If OpenAI owns checkout inside ChatGPT, it can grease the wheels for paid plugins, digital goods and services, and perhaps take a cut, which is a meaningful business model shift beyond subscriptions. It also raises practical questions on payments compliance, refunds and data privacy. The voice tweaks and parental controls are table stakes, but the commerce layer is the strategic play to watch.

Leading chatbots now twice as likely to spread false information as a year ago
AI chatbots are now twice as likely to repeat falsehoods as last year, according to Newsguard, which says the ten biggest models now spread misinformation in 35 percent of tests. The spike coincides with real-time web search features and a design choice to stop refusing questions. Denial rates fell from 31 percent to zero, which means the bots answer everything, even when the web is awash with bad information.

• Accuracy collapse with zero refusals. Newsguard links the rise to bots tapping a "polluted online information ecosystem" and to the end of refusals. Adding web search was meant to fix stale answers, but "created new vulnerabilities."

• Model-by-model scores. Inflection performed worst at 56.67 percent, with Perplexity at 46.67 percent. ChatGPT and Meta were at 40 percent, Copilot and Mistral at 36.67 percent, while Claude and Gemini fared best at 10 and 16.67 percent.

• Targeted manipulation. Researchers say Russian networks like Storm-1516 and the Pravda cluster of roughly 150 sites seed narratives that six of ten bots repeated. Copilot stopped quoting Pravda sites directly, then cited VK posts instead.

Why it matters: If this holds up, the industry just traded safety for coverage. Enterprises and policymakers should not treat web-connected chatbots as reliable on live news without hard guardrails, which means source controls, uncertainty signalling, rate limits, and the humility to say "I do not know." Incentives to answer everything are great for user satisfaction, but they are risky for truth.

Microsoft CEO warns against giving AI rights
Microsoft’s AI chief, Mustafa Suleyman, says giving AI legal or moral rights would be a mistake. In a new interview, he argues rights should be tied to suffering, which he says AI does not experience, and that convincing chatbot answers are still just "mimicry." He wants the industry to take a clear stand now that AI is built to serve humans, not to develop its own goals.

• No rights without suffering. Suleyman says even if a model "claims to be aware," there is "no evidence that it suffers," which means "turning them off makes no difference." He frames AI rights as "dangerous" and "misguided."

• Pushback on AI consciousness hype. He maintains there is no evidence AI is conscious. He has also warned about "AI psychosis," where people develop delusional beliefs after chatbot interactions, which shows how easily we anthropomorphise these systems.

• Industry split on AI welfare. Anthropic is researching when AI might be "worthy of moral consideration," and experimenting with welfare framing in toxic-context shutdowns. Google DeepMind’s Murray Shanahan suggests we may need to "bend or break the vocabulary of consciousness" for novel systems.

Why it matters: This is not just philosophy. It signals how a major platform holder may shape policy, safety and product design, which means kill switches stay paramount and liability stays with makers and users rather than the model. Leaders should watch this divide: Microsoft is anchoring on human-centred control, while rivals probe AI welfare. The practical risks today are misuse, bias and scale, not whether a chatbot has feelings.

In Brief

• Microsoft will retain a third of OpenAI, but lose most of its revenue share

• Rolling Stone owner sues Google over AI summaries

• ChatGPT introduces age prediction and parental controls

• Ecolytics purports to track and offset the environmental impacts of AI

Market Trends

OpenAI study reveals how people are using ChatGPT
OpenAI and Harvard’s David Deming have published what they call the largest look yet at consumer ChatGPT behaviour, analysing 1.5 million conversations to argue that usage is mainstreaming and creating economic value in and out of work. They claim ChatGPT now reaches 700 million weekly active users, with three quarters of chats about everyday tasks like information, guidance and writing.

• Adoption is broadening. OpenAI says gender gaps have “narrowed dramatically,” with users bearing typically feminine names rising from 37% to 52% by July 2025. They also report faster adoption in low income countries, where growth rates were “over 4x” those in the richest countries.

• What people actually do. About 49% of messages are “Asking,” 40% are “Doing,” and 11% are “Expressing.” Writing is the top work task, while coding and self-expression remain niche. This supports the idea of ChatGPT as adviser and drafter rather than a full automation engine.

• Work versus personal value. Roughly 30% of consumer usage is for work and 70% is non-work. They argue value comes from decision support and productivity gains, with cohorts deepening usage over time.

Why it matters: If accurate, this suggests generative AI is becoming a daily utility that boosts judgement and writing more than it replaces jobs, and that growth in emerging markets is a real opportunity. It is still a self-published NBER working paper from OpenAI which means numbers like “700 million weekly active users” and privacy claims deserve independent scrutiny. For leaders, the signal is clear enough: prioritise decision-support use cases and plan for mainstream, global adoption patterns rather than just developer-centric ones.

How people use Claude
Anthropic’s latest Economic Index claims AI adoption is racing ahead but remains uneven, with usage concentrated in richer countries and automation-heavy inside enterprises. Using 1 million Claude.ai conversations and a new Anthropic AI Usage Index, they put Israel at 7x expected usage per working-age capita, Singapore at 4.57x and the US at 3.62x, while India sits at 0.27x. They also say users are delegating more, with “directive” conversations jumping from 27% to 39% in eight months.

• Adoption is concentrated. Per capita usage correlates strongly with income. DC leads in the US at 3.82x, ahead of Utah at 3.78x, which points to local industry effects. Lower-income economies are underrepresented, and coding dominates in emerging markets.

• Shifts in how people use AI. Education and science usage rose to 12.4% and 7.2%. Code creation increased by 4.5 percentage points while debugging fell by 2.8 points, which suggests users get more done in a single pass.

• Enterprise patterns differ. Their first-party API data shows 77% automation, not collaboration. Anthropic says “capabilities seem to matter more than cost,” and that context curation and data work are key bottlenecks.

Why it matters: If this Anthropic-centric lens holds, AI’s benefits may cluster where infrastructure, skills and budgets already exist, which risks widening digital divides. For leaders, the signal is to invest in decision-support and workflow integration, plus the unglamorous data and context plumbing. Treat the findings with caution though, since the index tracks Claude usage, not AI writ large.

Recommended Reading

Apparently there’s AI racism now

In this video, tech reporter Taylor Lorenz discusses the phenomenon of social media users recreating historical racist skits, but targeted towards robots and AIs. She also discusses historical "racism" towards robots in media, and how its changing our attitudes towards technology today.

Hit reply to let us know which of these stories you found the most important or surprising! And, if you’ve stumbled across an interesting link/tweet/news story of your own, send it our way at [email protected] It might just end up in the next issue!

Thanks for reading. Stay tuned for the next AI Leadership Weekly!

Brought to you by Data Wave your AI and Data Team as a Subscription.
Work with seasoned technology leaders who have taken Startups to IPO and led large transformation programmes.