China mandates AI safety checks as Moonshot opens Kimi weights

China has introduced AI safety standards requiring domestic and foreign providers serving Chinese users to assess risks, protect user data and improve algorithmic transparency before launching models. The rules, launched on 3 November, set a February 2027 compliance deadline, with fines or exclusion from the Chinese market for violations. The scope of transparency requirements remains unclear, complicating planning for European firms already navigating the AI Act.

Moonshot AI is rolling out the open-weight Kimi K3.1 Pro, allowing developers to download its weights and host it themselves. The company claims improvements in coding, financial research and autonomous tasks. Self-hosting could help European organisations control sensitive data, but hardware requirements remain demanding, and independent testing is still needed.

Canada and Germany reaffirmed cooperation on AI sovereignty, safety and economic resilience. Discussions included investments in LawZero and support for Cohere following its merger with Aleph Alpha, but disclosed no funding amounts or detailed commitments. Quantum computing is also advancing through improved error correction, with forecasts pointing to hundreds of logical qubits by 2030 and thousands by 2035. Those projections remain uncertain.

Security researchers report rising attacks on Gmail accounts, including phishing and credential stuffing. Recommended defences include enabling multi-factor authentication, checking recovery details and refusing unexpected requests for sign-in codes. Users should also review connected apps and revoke unnecessary permissions, since malicious apps granted access can bypass the need to obtain a password.