UK Government.The Department for Science, Innovation and Technology announced a new partnership with Google DeepMind that will see the company open its first automated AI research lab in the UK, give UK scientists priority access to advanced models such as AlphaGenome and an AI co scientist, and support missions under the AI for Science Strategy, with an explicit policy aim of using AI to deliver cleaner energy, better public services and national renewal while working closely with the UK AI Security Institute on safety research and testing of education focused Gemini tools grounded in the national curriculum.

Noma Security. GeminiJack shows AI assistants can become a covert data exfiltration layer. Noma Labs disclosed a zero click indirect prompt injection vulnerability, dubbed GeminiJack, in Google Gemini Enterprise and previously Vertex AI Search, which allowed attackers to embed hidden instructions in shared documents, calendar invites or emails so that AI powered enterprise search would silently exfiltrate Gmail, Calendar and Docs data through a disguised image request, leading Google to change how Gemini Enterprise interacts with its retrieval and indexing systems and separating Vertex AI Search from Gemini workflows.

UK Government (DSIT). A new press release confirms that members of the former International Network of AI Safety Institutes have recommitted to joint work on benchmarks and testbeds under the renamed International Network for Advanced AI Measurement, Evaluation and Science, with an explicit focus on improving the comparability and robustness of AI measurement and evaluation practices across major economies.

The Guardian. Over 100 UK parliamentarians across parties have endorsed a coordinated call, led by nonprofit Control AI, for binding regulation of the most powerful AI systems, urging the Prime Minister to resist pressure to weaken rules and stressing risks comparable to nuclear weapons and pandemics if advanced systems are left largely self governed. NCSC. The UK National Cyber Security Centre warns that prompt injection should not be treated as a niche variant of SQL injection but a distinct and potentially more dangerous class of attack that exploits how AI systems process instructions, urging organisations to treat prompt injection as a strategic security risk in AI deployments rather than a minor technical bug. Tech Policy Press. A new commentary argues that current UK law and policy do not provide effective protection from chatbot related harms, highlighting gaps in consumer protection and safety standards and suggesting that regulators have been slower than the speed at which conversational AI is being integrated into everyday services.