Nobody can blame the software

AI-Drafted Safety Policy Fails as Legal Defense

WorkSafeNB, the workplace safety regulator in New Brunswick, Canada, warned on September 10 that its consultants keep finding employers relying on AI to write safety policies, procedures and codes of practice. The drafts cited the wrong sections of the law, used old standards and mixed up Canadian and American rules. One consultant had a tool invent a book as its source. Fall protection, trenching and hot work are where such errors do the most harm.

For managers: the regulator's consultant said AI "would not in any way ever be able to be used as a legal defense," so whoever signs the procedure has to check every clause it cites.

Source: thesafetymag.com

San Mateo Moves to Make Employers Plan for Jobs Before Using Robots

Supervisors in San Mateo County, California, voted 4 to 0 to advance permits for autonomous robots, with a final vote to come. The rule covers machines that use cameras or sensors and work at least partly without human control near workers or the public. Applicants must file an economic impact assessment and a workforce plan. Displaced staff get redeployment at equal or better pay, or 60 days' notice with severance, or the employer pays into a retraining fund.

For managers: wherever you operate, this is the right order: decide what happens to the people before the machine is switched on.

Source: ohsonline.com

Chipotle Using Palantir to Score Each Restaurant's Food Safety Risk

Chipotle is piloting a food safety risk platform built on Palantir's software, according to screenshots seen by Wired. It appears to combine each restaurant's health department scores, pest incidents and employee illnesses into one score for the store's risk level. Chipotle says it gives teams one consistent view so they can prioritize follow-up. How widely it runs is not known. The United States has recorded more than 12,800 illnesses from foodborne outbreaks this year.

For managers: if staff illness feeds the score, make sure a crew member who reports being sick never counts against the store, or the reports will quietly stop.

Source: wired.com

One number

64%

of hourly workers whose employers already use AI report a positive impact, yet half of all hourly workers still worry AI will replace their jobs. That comes from a Harris Poll survey of 2,208 hourly workers aged 18 to 45 in the United States, run for DailyPay and Workday. The fear is about AI in general; the verdict from people already working with it is mostly positive.

Try this Monday

Pull out one safety procedure, policy or checklist that was drafted or updated with AI help. Check every section of the law and every standard it cites against the original. Put the checker's name and the date at the bottom. If nobody on site can, find someone qualified before it takes effect.

Worth a look

Software that disciplines. An Organisation for Economic Co-operation and Development survey of 6,047 firms found 67% of American firms use software to sanction poor performance, against 4% in the four European countries surveyed.

The people behind the agent. Meta's Muse assistant hands some of its phone calls to contractors in call centers, and some testers were told only afterward.