AI & Society Brief — October 2026

This month, the rules caught up with decisions already being taken. A person must remain the one who decides, an employer must say when AI played a part in a layoff, and someone must answer when a system harms a child or copies a voice. Meanwhile, workers are using AI more while getting less help to learn it.

Italy makes a human the only lawful decision-maker in AI-assisted hiring and firing

• Source: ActuIA · 19 October 2026 · actuia.com — based on analyses by EcNews (Euroconference) and A&O Shearman, 13 October 2026
• Italy adopted Legislative Decree No. 179 on 7 October 2026 and published it in the Gazzetta Ufficiale (No. 234) on 8 October. It aligns national law with the EU AI Act and applies from 23 October 2026. Under its labour provisions (Article 41), decisions on hiring, changes to a contract, dismissal, disciplinary sanctions and performance reviews can no longer rest solely on automated processing. The final decision must come from a person with "effective and autonomous" authority. Employees must be informed before processing and can ask for an intelligible explanation, including how much weight the system carried.
• A decision that breaks the rule is void, and the employee's consent does not cure it. CV screening and candidate search do not count as final decisions.
• Why it matters: The rule applies now, even though the matching EU obligations on high-risk HR systems have been pushed back to December 2027. Any organization with staff in Italy needs to be able to name, for each AI-assisted HR decision, the person who actually decided and what they based it on. A human signature on a system's output will not be enough.

In California and Connecticut, employers must now say when AI is behind a layoff

• Source: TechRadar · October 2026 · techradar.com — Bloomberg Government · news.bgov.com — CDF Labor Law · 5 October 2026 · cdflaborlaw.com — Ogletree Deakins · ogletree.com
• On 30 September 2026, California's governor signed two laws:
  • SB 947, operative 1 July 2027, bars employers from relying solely on an automated decision system to discipline or dismiss someone. A human must review the decision and corroborate it, and the worker must receive written notice and a human contact. A broader version was vetoed in 2025.
  • SB 951, effective 1 January 2027, requires mass-layoff notices driven "in whole or in substantial part" by AI to state the number, location and types of jobs displaced and the technology involved. The state must publish summaries of these notices and a quarterly report.
• In Connecticut, Public Act 26-15 has applied since 1 October 2026. Employers filing a mass-layoff notice must tell the state labour department whether AI or another technology contributed. The use of an automated hiring tool is also no defence against a discrimination claim.
• Why it matters: Until now, saying a job cut was caused by AI was a communication choice. In two US states, it is becoming a declaration that is public, recorded and enforceable. Organizations that cite AI to justify restructuring, or avoid citing it, should expect that wording to become a record that others can check.

China's Supreme Court tells judges how to rule on deepfakes, voice cloning and personalised pricing

• Source: Reed Smith · September 2026 · reedsmith.com — CMS · cms.law — LexisNexis, 9 September 2026
• On 7 September 2026, China's Supreme People's Court issued its first opinions on how courts should handle AI-related disputes. It is guidance that courts apply alongside existing law, not new legislation. It treats creating an identifiable image of a person without consent, including face-swapping and voice cloning, as a possible breach of portrait, name or voice rights, and courts may grant urgent injunctions. Using AI to piece together someone's private information from public sources ("cyber doxxing") is treated as a privacy violation. Price discrimination based on user profiling that harms consumers makes the service provider liable in tort.
• The guidance sets fault-based liability as the general rule. It also allows public personal data to be used for model training within a "reasonable scope", unless the person has explicitly objected.
• Why it matters: The everyday harms people face from AI now have a clear legal route in the world's second-largest consumer market: a cloned voice, a stolen face, or a higher price because of a profile. Any household or business there that has been targeted by impersonation or opaque pricing has a better-defined basis for action than it did a month ago.

California sets default limits on how long children can talk to companion chatbots

• Source: Common Sense Media (bill sponsor), press release · September 2026 · commonsensemedia.org — Akin Gump · akingump.com
• On 10 September 2026, California's governor signed SB 1119 ("Adam's Law"), named after a teenager who died in 2025. From 1 July 2027, operators that allow children on companion chatbots must switch on default parental controls. These controls turn off long-term conversational memory and push notifications, and cap sessions at one hour and daily use at two hours. Operators must also have crisis-response protocols and accept restrictions on advertising and data use.
• Prosecutors can seek up to $5,000 per affected child for negligent violations and $15,000 for intentional ones. Harmed children or their parents can sue. Common Sense Media, which publicised the signing, campaigns for children's online safety and backed the bill.
• Why it matters: The law regulates how these products are designed, not what they say: memory, notifications and session length are the features that keep a child coming back. Parents and schools elsewhere can use the same three settings as a practical checklist today, without waiting for local rules.

Workers' use of AI is rising while their access to training is falling

• Source: PwC, Global Workforce Hopes and Fears Survey 2026 · 29 September 2026 · pwc.com — Benefits and Pensions Monitor · benefitsandpensionsmonitor.com
• PwC surveyed 49,364 workers in 48 countries and regions and 29 sectors in May–June 2026. 64% had used AI at work in the past year, up 10 points on 2025, and daily use of generative AI rose from 14% to 22%.
• Over the same period, the share who say they can access the learning and development they need fell from 59% to 51%. Among the 56% of workers PwC calls the "engine room", only two in five have that access. 29% of this group say they are very or extremely likely to change employer within a year.
• PwC sells AI and workforce consulting, so it has an interest in how visible the issue is.
• Why it matters: The gap is not between people who use AI and people who don't. It is between use and support, and it is widening. Organizations that cut training budgets while their staff's AI use rises are leaving people to learn on their own, and the survey links that to the intention to leave.