Introduction
This fortnight is defined by (1) new UK international AI and science partnership announcements, (2) a clear shift from guidance to enforcement under the Online Safety Act, and (3) EU implementation work (sandboxes) alongside the Digital Omnibus simplification track that UK EU-facing providers must monitor.
Executive snapshot
- DSIT (see also AISI): UK announced a new partnership with Google DeepMind to translate AI into science, energy and public-service gains, with a published MoU on AI opportunities and security.
- GOV.UK: UK Tech Minister’s Canada visit produced fresh international agreements on AI and digital government, plus a G7 deal focused on SME AI adoption.
- Ofcom: Ofcom issued its first £1 million Online Safety Act fine against an adult website provider for inadequate age checks, plus a further penalty for failing to respond to information requests.
- MoD (see also UK Parliament, Royal Navy): MoD launched Atlantic Bastion, explicitly combining autonomous vessels and AI to protect undersea infrastructure, with early Parliamentary scrutiny.
- EU Commission: The Commission opened feedback on a draft implementing act on AI regulatory sandboxes under the AI Act (closes 6 January 2026).
1. UK policy and international positioning
- The DeepMind partnership is framed as delivering science breakthroughs and smarter public services, supported by a formal MoU on AI opportunities and security.
- The UK AI Security Institute confirms a new DeepMind research partnership as part of DSIT’s wider collaboration narrative.
- The Canada visit announcement ties AI to public-service modernisation and SME adoption through G7 partners, signalling a practical “adoption agenda” alongside safety rhetoric.
2. Regulator action
- Ofcom: Ofcom’s £1m fine against AVS Group shows Online Safety Act compliance is now moving into high-visibility enforcement, not just guidance.
- Ofcom: The enforcement notice links the fine to the new legal duty to prevent children encountering pornographic content using age assurance, and adds a separate penalty for non-cooperation with information requests.
3. National security and defence adoption
- MoD: Atlantic Bastion explicitly describes an “AI” and autonomous-vessel hybrid force to protect undersea cables and pipelines, anchored to heightened Russian undersea activity.
- The Chair of the Joint Committee on the National Security Strategy publicly responded to the “early details” of Atlantic Bastion, signalling immediate governance scrutiny.
- The Royal Navy frames Atlantic Bastion as an advanced hybrid naval force for tracking and responding to adversaries across large ocean areas.
4. EU developments that affect UK EU-facing systems
- European Commission: The Commission is collecting feedback on rules for establishing and operating AI regulatory sandboxes under the AI Act, which matters for providers planning EU testing pathways.
- European Commission and Reuters: The Digital Omnibus on AI proposal is presented as targeted simplification for smoother AI Act implementation, and Reuters reports it would delay “high-risk” AI rules to December 2027.
5. Continuing items and deadlines to track (kept short)
- DSIT: AI Growth Lab call for evidence remains open, with the stated purpose of enabling targeted regulatory modifications under safeguards (deadline 2 January 2026).
- ICO: ICO enforcement procedural guidance consultation remains open (deadline 23 January 2026).
Conclusion
This fortnight’s real shift is the move from policy narratives to operational levers: international AI partnerships and adoption deals on the one hand, and sharper regulator enforcement and security-driven AI deployment on the other. For UK organisations, EU sandbox rules and the Digital Omnibus track together mean compliance planning must stay agile for EU-facing systems while UK enforcement expectations become more concrete.
Sources: GOV.UK, DSIT, ICO, Ofcom, Ministry of Defence, UK Parliament Committees, Royal Navy, European Commission, AISI