Daily Update, 13 October 2026: UK to put frontier AI to work on government cyber defences

The UK government on Monday 12 October set out how it plans to strengthen Britain's defences against AI-enabled cyber risks, in a speech by AI minister Kanishka Narayan and a gov.uk press release from the Cabinet Office, the AI Security Institute, the National Cyber Security Centre and DCMS. Also today: Meta's Muse Wonder models, the AI Security Institute's priorities for frontier-AI best practice, Nvidia's rack designs for the Open Compute Project, a combined Grok, Cursor and X plan, and the IMF on AI share prices as reported by the Guardian.

UK to put frontier AI to work on government cyber defences

Speaking at Imperial College London's Base One incubator, Narayan said: "I want Britain to become one of the hardest countries in the world to attack using artificial intelligence." The press release lists these measures:

• Frontier AI for cyber defence. The government is extending a programme that the release says builds on two pilots which "have already helped identify and fix over a thousand potential cyber vulnerabilities across government systems". In the speech, Narayan said the central vulnerability-monitoring service "has already helped reduce the time taken to fix identified weaknesses from 50 days to just eight".
• The Cyber Security and Resilience Bill. The government "will explore strengthening parts of the Cyber Security and Resilience Bill and the wider regime" so that essential services such as hospitals and utilities are better protected, and will look at how near-miss incidents can be identified and reported. The release says the detail "will be worked through with parliamentarians", technical experts and operators.
• Incidents involving AI agents. The release says work is under way with the NCSC and the AI Security Institute on how such incidents should be responded to. The release says a major incident involving AI agents "could play out quicker and wider" than other major incidents governments are designed to handle.
• AI verification. More details on support for UK AI verification work "will be set out in the coming months".

The release calls the package "the first steps on a longer journey". We cover the speech in more detail in a separate piece.

AISI sets out its priorities for frontier-AI best practice

The AI Security Institute (AISI) published a blog post on 12 October setting out "the technical practices most urgently needed" by frontier AI developers. It names two new challenges: risks "arising earlier in the development process", including during training, internal deployment and evaluation, and evaluation practices that are "struggling to keep pace". It lists eight priorities, which it says are not exhaustive, including detecting, responding to and reporting agentic AI incidents; sharing information on the speed of AI development; and enabling independent auditors and evaluators. AISI says it "is not seeking to mandate requirements for frontier developers" and will work with developers to develop voluntary practices. We cover the full list of AISI's priorities in a separate piece.

Meta releases Muse Wonder

Meta on 12 October introduced Muse Wonder and Muse Wonder Full, which it calls "our most capable models" and says "deliver state-of-the-art performance on personal and professional agentic tasks". TechCrunch described the release as Meta's first attempt at a frontier model. Meta says Muse Wonder "is available today through the Meta Model API and Muse Code", and Muse Wonder Full is available to Muse Code subscribers. According to TechCrunch, Muse Wonder costs $4 per million input tokens and $20 per million output tokens on the API, and Wonder Full will not be served through the API. Meta says the models were trained at its Prometheus data centre and that "A full report with additional safety evaluations will be available soon". Meta's blog compares the models with other frontier models in benchmark charts; we have not checked those figures.

Nvidia contributes Vera Rubin rack designs to the Open Compute Project

At the Open Compute Project (OCP) Global Summit in San Jose, Nvidia said on 12 October that it had contributed "key NVIDIA Vera Rubin rack and compute tray designs to OCP through the open NVIDIA MGX architecture". It says the updated MGX 1.2 specification "supports rack deployments up to 250kW" and adds a liquid-cooled 5,000A busbar. Separately, Nvidia and Cisco announced plans for USFP, a pluggable optics module that the companies say "is designed to be the first module to support 400Gb/s-per-lane electrical signals". They say it delivers four times the bandwidth density and 1.3 times the power efficiency of 200Gb/s-per-lane optics, and that they plan an open industry specification through an upcoming community-led multi-source project, and are working with TE Connectivity and Amphenol on cages, connectors and modules.

Grok, Grok Bot, Cursor and X Premium on one plan

The company behind Grok, which brands itself SpaceXSI on x.ai (the page footer names SpaceXAI LLC), introduced "The X Subscription" on 12 October, bringing "Grok, Grok Bot, Cursor, and X Premium in one plan". It says it will make the plan available "to everyone globally" over the next few days. The pricing table has five rows: Free ($0), Plus ($8 a month), Premium ($30), Super ($100) and Ultra ($200). The company says Premium includes "The benefits of SuperGrok, Cursor Pro, and X Premium+, previously $90/month combined". Paid plans share a pool of Grok model usage across Grok and Cursor; third-party models in Cursor remain available on Premium and above, "billed separately at API list prices". Prices on the page are in US dollars.

IMF flags an AI share correction as a risk to financial stability, the Guardian reports

According to the Guardian, the International Monetary Fund's twice-yearly Global Financial Stability Report names an AI correction as one of three "salient risks" to stability, alongside rising government borrowing costs and the threat of a big cyber-attack. The Guardian quotes the report as saying: "The consequences of an AI correction could be global". It reports that, aside from the US, Asian markets would be most exposed, with the UK and euro area potentially more insulated, though their investors with exposure to the US would be hit. imf.org returned "Access Denied" when we tried to read the report, so we have not seen the IMF's text; the quotes here are as the Guardian reported them.

Why it matters

On our reading, the two UK publications on 12 October point the same way on one issue: the government says it is working out how to respond to incidents involving AI agents, and AISI lists detecting, responding to and reporting agentic AI incidents among its priorities for developers. How either turns into practice depends on detail the government says will be worked through with Parliament and experts, and on AISI's voluntary work with developers.