The easy half is buying it

Ten thousand cameras now watch for unsafe work

Tata Steel has put more than 10,000 cameras with AI across its plants, trained to spot unsafe behaviour and alert a supervisor while the job is still happening. Gate passes and beacons flag anyone in an area they should not be in. Remote operation centres pull camera feeds and machine data into one room so fewer people stand near the hazard. Truck cameras watch drivers for signs of fatigue and phone use.

For managers: the point is not a better accident investigation, it is an interruption while the job is still running, which only works if a supervisor is free to act.

Source: hrkatha.com

Papa Johns replaced the old system before trusting the forecast

Papa Johns is rolling out new point of sale and back office software from PAR Technology across 3,200 restaurants in the United States, replacing a system the company wrote about thirty years ago. Managers now use AI forecasting to predict sales and plan labour against it. The digital chief, Kevin Vasconi, said the new screens look like an app rather than walls of text, and that training is faster and order accuracy better.

For managers: a forecast is only as good as the system feeding it, so if scheduling runs on whatever an old machine happens to collect, fix that first.

Source: restaurantdive.com

Most hotels now use AI, few get the time back

The State of Distribution 2026 report, from RateGain with New York University's hospitality school and the hotel distribution body HEDNA, covers 270 hotel brands and more than 58,000 properties. More than half use or are buying generative AI. Fewer than one in ten cut manual work by more than 30%. The reasons given are scattered data and reluctance to let software decide alone. Over 80% of commercial teams still spend a day or two each week on manual reporting.

For managers: buying it is the easy half, and the report's own explanation is that the work was never rebuilt around the tool.

Source: hotelnewsresource.com

Britain asks who should check the software that scores workers

The British government is consulting on new rules for monitoring workers, and the deadline is 30 September. Three options are on the table: a code of practice that tribunals could use to adjust compensation by up to 25%; a duty to consult staff or unions before bringing in any monitoring tool; or voluntary guidance with no penalties. The Information Commissioner's Office says one in three British organisations now monitor digital activity, up from one in five three years ago.

For managers: the regulator's line is that signing off an automated output without examining it may not count as human involvement, so write down who reviews your rosters and scores, and what they check.

Source: peoplemanagement.co.uk

Korea will translate safety briefings into workers' own languages

The Korea Occupational Safety and Health Association signed an agreement on 8 September with Flitto, a translation company, to put live machine translation into the safety work that happens on site: risk assessments, toolbox talks at the start of a shift, work permits and accident reports. The languages are those of countries sending workers under Korea's employment permit scheme. The two will build shared glossaries of hazard and safety law terms, offered through 28 regional branches.

For managers: if the toolbox talk, the permit and the incident form are in a language part of the crew barely reads, what you hold is a signature rather than an understanding.

Source: venturesquare.net

One number

42%

of workers who use AI at work say they have gone ahead with an answer they suspected was wrong. That comes from 500 employed adults in the United States, surveyed on 1 September by Kolmogorov Law, a California firm. The same survey found 22% work somewhere with a written rule that AI output must be checked, and 59% say no such rule exists. The behaviour is not the finding. The empty policy shelf is.

Try this Monday

Pick one thing AI now drafts or decides for your team: a roster, a demand forecast, a safety alert. Ask who looks at it before anyone acts, and whether that is written down anywhere. If it is not, one line in a procedure fixes it this week.

Worth a look

Singapore's agent managers. Agentic AI runs at 51% of organisations, up from 22% a year ago, while 10% have redesigned a process end to end.

A gap worth measuring. 58% of healthcare executives rate their leadership effective, against 28% of frontline staff, across 4,040 responses.

California's deadline. The governor has until 30 September to sign or veto the remaining AI bills, including employment and workplace surveillance measures.