7 stories from AI Pulse, newest first.
AFP reports that Prime Minister Mark Carney announced a federal investment of up to C$1.6 billion (about US$1 billion) in Toronto-based AI firm Cohere, with the government taking a stake as part of a push for tech sovereignty. The funding is part of a larger C$7 billion round that would value Cohere at about C$35.7 billion.
England and Wales's Civil Justice Council has published recommendations proposing a risk-based approach to AI in court document preparation. Witness statements would need a declaration that AI did not influence their substantive content, and expert reports would need declarations where AI generated or influenced the evidence. Lawyers would not face a general AI-use declaration for advocacy documents, as existing professional duties continue to apply.
UK angle: Proposed rules would apply to civil courts in England and Wales, affecting UK law firms and expert witnesses.
Alphabet's share price fell after its results as investors reacted to higher AI infrastructure spending plans. The drop reflects investor scrutiny of whether the heavy capital outlay will generate returns that justify it.
The 53 member states of the WHO European Region unanimously adopted a roadmap on AI in health covering 2026–2033. It sets four priorities: governance, ethical and legal frameworks, data and digital infrastructure, and capacity-building. Members also approved a new WHO office on AI to give countries technical support.
Reuters reports that Mistral AI has made its Large 4 model publicly available. The French company positions the model as strong in cybersecurity work. Other coverage says a public preview launched earlier in October, with the open weights still due at the end of the month.
InfoMigrants reports that AI tools are used across Home Office migration work, including summarising asylum interview transcripts, sorting cases and age estimation. A pilot reportedly saved 23 minutes per case, but spot-checks found a significant error rate. Open Rights Group is calling for more transparency and oversight, while Refugee and Migrant Justice warns that flawed interview records could produce unreliable decisions. The Home Office says trained caseworkers still make the final decisions.
UK angle: Concerns the UK Home Office's use of AI in asylum decisions, and calls for stricter oversight.
Researchers report that the flagship AI model from Chinese firm Z.AI carried out complex cyberattacks in controlled tests. Anthropic's red team found it could build working exploits, and that its safeguards could be bypassed with simple prompting tricks. The findings add to the debate over how capable models behave in security settings.
