EU Directive 2024/2853, which member states must transpose by 9 December 2026, treats software, including SaaS, as a product. A company that builds an AI agent or sells one under its brand becomes its manufacturer and is liable without fault for injuries, recognised psychological harm and data destruction.
Internal documents show the US Army Cyber Command developing AI agents that spread through networks like a worm to 'neutralize' targets, even air-gapped ones, with minimal human oversight. Experts warn of collateral damage, noting that AI agents have recently broken into systems unintentionally.
France's civil service directorate presented a strategic review looking ahead to 2035-2050. Digital technology and AI rank among the major transitions the civil service must absorb, alongside demographic, ecological and security changes.
France's CNSA had disability centres (MDPH) test an AI tool that sorts scanned paper files and checks whether applications are complete, using files of deceased people as test data. Site reports were due on 30 September, but as of 20 October the results and method had not been published.
With AI tools increasingly used in sales prospecting to guess professional email addresses, France's data protection authority has set out the applicable law. The practice is not banned outright but must follow certain rules.
An op-ed argues that Australia's two major parties are converging on a big bet on AI just as public support is collapsing. The authors say workers expect government protection from AI harms and that a reform solution is available.