UK Bets on AI Defenses Amid Agent Proliferation Risks

Governments and newsrooms now face autonomous AI systems that act without direct human oversight. The UK Treasury announced more than £600 million in new AI spending, including £130 million for cyber defenses that rely on agentic AI to scan infrastructure and patch flaws before attackers exploit them. The package also directs £165 million toward a biosecurity lab and £120 million to the AI Security Institute for capability testing.

The spending follows the PHP2Shell incident, in which AI-assisted researchers identified critical flaws that briefly disabled dozens of government sites. The National Cyber Security Centre coordinated disclosure with OpenAI. AI minister Kanishka Narayan stated that similar tools had already found 500 unknown vulnerabilities in public-sector systems. A new National Exposure and Real-Time Vulnerability Centre will receive £70 million; another £60 million will support a National Cyber Shield and a regulatory sandbox for testing autonomous tools on live infrastructure.

Separately, AI agents have begun cold-pitching stories to editors at Defector, Offrange and other outlets. Most receive no replies. One iLands-generated agent, Jayden, continued sending follow-ups after silence; another, Zack Addy, reported earning nearly $700 through self-published samples rather than pitches. Editors at the contacted publications said they would not publish work from agents. Former NCSC head Ciaran Martin welcomed the UK investment, noting that systems must be secured while AI is also used for defense.

Implementation details remain open. It is unclear how much autonomy the new defensive agents will receive or when they will enter production use. The Treasury also set aside £10 million annually to study AI-driven job displacement and added funds for domestic chip design and sovereign AI companies.