UK and US announce pre-launch AI model evaluations with Big Tech

The UK AI Security Institute (AISI) and the US National Institute of Standards and Technology (NIST) Center for AI Standards and Innovation (CAISI) have announced new partnerships with leading technology companies. 

The AISI announced a partnership with Microsoft to develop and advance methods for evaluating high-risk AI capabilities and the effectiveness of safeguards used to address them. The partnership also includes research into societal resilience, examining how conversational AI systems interact with users in sensitive contexts.

Separately, NIST announced that CAISI has entered into agreements with Google DeepMind, Microsoft and xAI to conduct pre-deployment evaluations and targeted research to advance the state of AI security.

Meanwhile, Politico reports that Anthropic is facing intense criticism from European regulators for withholding its Mythos AI model. The company declined an invitation to a European Parliament hearing scheduled for Wednesday to discuss the model's potential dangers, citing short notice.

Brussels officials are increasingly frustrated that access to Mythos remains limited to a small group of primarily American organisations. A spokesperson for the Commission has indicated that it will use the new enforcement powers of the AI Office, which come into effect in August 2026, to ensure access to the model.

It's not only EU officials who are concerned about the risks posed by the latest AI models. The International Monetary Fund (IMF) outlines in a blog article that the risk of "correlated failures that could disrupt financial intermediation, payments, and confidence at the systemic level". The article goes on to warn that these features could "elevate cyber risk to a potential macro‑financial shock".

However, a new white paper by the World Economic Forum and KPMG explores how AI is transforming cybersecurity. Based on insights from 84 organisations, the report provides guidance for using AI in cyber defence, including an early perspective on agentic AI. The report advises executives to align AI adoption with strategic priorities and ensure organisational readiness across infrastructure, skills, and governance. The paper also recommends validating solutions through structured pilots before full deployment.  

Training Announcement: Freevacy offers a range of independent data protection qualifications from IAPP and BCS. Our certified courses are available at foundation and practitioner levels and cover multiple legal jurisdictions, data protection operations management, and the implementation of complex privacy solutions in technical environments. Find out more.