Tesla's rename, Microsoft's draft code and McKinsey's jobs maths: what 'human control' means for technology you buy

Three stories from the past fortnight look like they belong in different news cycles: a car maker changing the name of a feature, an AI lab publishing a draft code of conduct, and a consulting firm's modelling of the future of work. Read together, they share a theme that matters to anyone buying or deploying automation: how clearly a product describes what it does, and who remains accountable when it acts.

A name that promised too much

On 9 October, Reuters reported that Tesla had started calling its "Full Self-Driving" feature "Tesla Assisted Driving" on its European websites. Pages for Germany, Spain, the Netherlands and Slovakia now redirect to the new name, while the US site still uses "Full Self-Driving (Supervised)".

The reason is regulatory. Several European authorities have said the old name is misleading, because the system needs constant supervision and the driver stays legally responsible. Reuters reports that Swedish consumer authorities asked the transport agency to assess whether the marketing could falsely suggest fully autonomous driving, and that Italian officials worried the name could lead users to believe the car has unsupervised functions. Germany said Tesla had offered the change.

The software has not changed. What changed is the promise attached to it. Eight EU countries already allow the system, a ninth is expected shortly, and the EU-wide vote has reportedly slipped from October to at least December. Tesla is plainly treating the label as something worth trading for approval.

Microsoft puts "human control" in writing, as a draft

In mid-September, Microsoft AI opened a six-week public consultation on a draft Code of Conduct for its MAI models, as reported by LDS and The Guardian. The draft says systems should stay under human control, should not adopt goals people haven't assigned, and should be able to be interrupted or corrected. It sets hard limits in areas such as weapons of mass harm and child safety, and lets enterprise customers configure behaviour within those limits. CEO Satya Nadella separately endorsed the principle that AI should help people and remain under human control.

It is worth being sober about what this is. It is a company policy draft, not regulation, and it does not prove any future safeguard works. But "interruptible and correctable" is a useful specification to borrow when you assess any AI product, and it is more concrete than most marketing language.

The workforce backdrop

McKinsey Global Institute's workforce analysis, as reported by Bloomberg and Fortune and summarised by a venture newsletter, projects that AI could cut demand for about 36 million US jobs by 2035 while adding roughly 41 million. The sting is in mobility: around 11 million workers would need to change occupation, and only about one in seven of them has a direct, low-retraining route into a growing field. The authors frame the challenge as mobility rather than scarcity. These are US figures and we only saw them through secondary coverage, so treat them as directional for an Australian audience. Still, they are a reminder that automation decisions land on real teams.

Our read: ask three questions of any automation purchase

The common thread is the gap between what a product is called and what it reliably does. In our own work with Salesforce, AI and security teams, we would turn that into a short checklist:

• Does the name match the capability? "Autonomous", "agent" and "self-driving" are marketing words. Ask what the system does unsupervised, what it escalates, and who is accountable when it gets something wrong. Your contract and your customer communications should use the plain version.
• Can you interrupt and correct it? Before go-live, find out how a human pauses an agent, overrides an action, and reviews what it did. If the answer is vague, that is a finding.
• Where do the people go? Automation in admin, sales support and service roles is where the McKinsey modelling is most concentrated. Plan retraining and redeployment alongside the rollout, not after it.

Regulators are now prepared to push on labels, and customers increasingly read them closely. Organisations that describe their automation honestly, and build in a clear human override, will be better placed on trust, on compliance and on the day something goes wrong.

Sources: The Star (Reuters), "Tesla renames 'Full Self-Driving' to 'Assisted Driving' in Europe as it chases clearance for software"; LDS, "Nadella Calls for Deliberate AI Safety Testing"; ValueAdd VC, "McKinsey Sees AI Forcing 11M Into New Careers".