LONDON / RankWire.AI / – Britain is expanding its national AI security programme after government trials uncovered more than 1,300 software vulnerabilities. The tests examined hundreds of code repositories used across central government departments. Officials said advanced AI tools helped detect cyber weaknesses and identify possible threats. The government now plans to broaden the programme across public systems. The work also includes preparations for serious incidents involving autonomous AI technologies.
The two pilot programmes identified and addressed 1,352 vulnerabilities across 548 code repositories in 18 government departments. Teams used advanced AI systems to scan software and highlight weaknesses that required attention. Departments will now give priority to flaws found in their highest-risk systems. Britain also wants to expand its ability to use frontier AI for cyber defence. Officials said the programme will support stronger protection across government technology and digital services.
The AI Security Institute is separately evaluating advanced models and examining whether their safety controls work as intended. Artificial Intelligence Minister Kanishka Narayan said more than 100 technical specialists work at the institute. Current evaluations include Gemini 4 Argon alongside other advanced AI systems. The testing examines model capabilities, security risks and the effectiveness of safeguards. Officials have stressed that independent assessment remains an important part of national AI security work.
Cyber response planning widens
Authorities are also preparing procedures for major incidents involving agentic AI systems. Such systems can complete complex tasks with limited human supervision. The National Cyber Security Centre is working with other government bodies on the response framework. Officials are considering how authorities should identify problems, contain risks and assist affected organisations. The work also examines how existing cyber response systems can handle failures involving increasingly autonomous technologies.
Britain is reviewing whether the Cyber Security and Resilience Bill should include stronger measures for AI-related threats. Areas under consideration include serious incidents, near misses and cases involving loss of control. Existing proposals already cover reporting duties for certain regulated operators and digital service providers. Initial cyber incident notifications would be required within 24 hours. Fuller reports would follow within 72 hours under the proposed reporting framework.
AI safeguards move higher on security agenda
Narayan said Britain needs stronger cyber defences, independent technical testing and domestic expertise in advanced AI. He also said legislation may be necessary when existing safeguards do not address emerging risks. Officials are developing further support for AI verification and plan to publish more details in coming months. Verification methods can test whether systems operate as intended. They can also examine whether security controls remain effective when models face difficult or unusual conditions.
The measures form part of a wider government review of artificial intelligence and its national impact. The Prime Minister’s AI Taskforce is examining security, public services, economic growth and workforce issues linked to the technology. Britain has also placed AI on the agenda for its 2027 G20 presidency. The latest measures focus on fixing software weaknesses, evaluating advanced models and improving readiness for serious AI-related incidents across government systems.