The UK is funding a new AI Biosecurity Lab with £100 million of government money and up to £200 million of OpenAI and Anthropic model credits
AspireAI community · UK Cabinet Office and HM Treasury (GOV.UK)
The Cabinet Office announced on October 24 that it is establishing the AI Biosecurity Lab, an independent non-profit modelled on the UK's AI Security Institute, with £100 million of government funding and up to £200 million in commitments from frontier labs: OpenAI provides £100 million in compute and model credits plus trusted access through its Rosalind Biodefense programme, and Anthropic up to £100 million in model credits, trusted model access and technical support, with Google DeepMind and Basecamp Research joining as research partners. The stated mission is early detection of emerging biological threats and faster development of vaccines and medical countermeasures, and the wider package includes £165 million for a National AI Biosecurity Initiative, an extra £120 million for the AI Security Institute, £60 million for a National Cyber Shield that uses agentic AI for cyber defence, and £70 million for a National Exposure and Real-Time Vulnerability Centre. The honest limitation is shape rather than ambition: most of the headline backing is model credits and access rather than cash, and the government release itself does not state a package total, a timescale for most lines, or who will lead the lab or sit on its scientific advisory committee.
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.