Blog
The Execution
The cybersecurity industry has spent decades and trillions of dollars building tools that operate after the fact.
Detect the breach. Respond to the incident. Recover the systems. Investigate the fraud.
Every one of those disciplines assumes the damage has already occurred and the goal is to limit it. Below, we examine what that assumption costs: real breaches, real losses, real organisations that did everything the industry told them to do and still lost. The question running through every post is the same: at the moment the consequential action executed, where was the human who should have stopped it?
DeepMind says high-risk actions need real-time blocking, not after-the-fact review. They have not answered who does the blocking.
Google DeepMind published its AI Control Roadmap this week, securing its own internal systems against advanced, imperfectly aligned AI agents. Buried in the framework is a distinction the wider industry has been avoiding: most agent actions can be reviewed after the fact. The high-consequence ones cannot. DeepMind calls this asynchronous versus synchronous response. They have built the case for it. They have not built the mechanism that makes it real.
Read articleThe NCSC's warning is clear. Vulnerabilities you tolerate today will be exploited in conflict tomorrow.
Richard Horne, CEO of the National Cyber Security Centre, delivered the RUSI Annual Security Lecture on 17 June 2026. Three quarters of cyber attacks on UK critical infrastructure are linked to hostile states. AI will accelerate exploitation of known vulnerabilities by 2028. The vulnerabilities tolerated today will be exploited in conflict tomorrow. The commercial argument for securing your critical assets before working the patch backlog has never been stated more clearly by a government source.
Read articleThe FBI recorded $893 million in AI-related scams in 2025. The primary method was executive impersonation to approve payments.
The Soufan Center published its assessment of malicious AI use on 17 June 2026. The FBI figure it cites is not an abstract threat statistic. It is a description of a specific attack against a specific control point. That control point is the execution boundary of a high-consequence financial action. It is unoccupied in most organisations.
Read articleThe patch cycle cannot win a race against AI-accelerated exploitation. There is a different starting point.
The market for vulnerability scanning, attack surface management, and penetration testing is saturated and commoditised. Frontier AI can now find zero-day vulnerabilities at scale in hours. The patch cycle cannot keep pace. It is time to invert the problem.
Read articleAll Ten OWASP Agentic AI Risks Are Now Real Incidents. The Same Control Would Have Stopped Them.
The OWASP State of Agentic AI Security and Governance v2 is 139 pages of documented incidents, unsolved problems, and emerging patterns. One control appears in the unsolved problems section. It is not unsolved
Read article54% of security professionals have faced an AI-related incident. Same root cause.
A Check Point survey of over 1,000 cybersecurity professionals found that more than half had experienced an AI-related security incident. Unauthorised AI usage. AI-generated attacks. Sensitive data leakage. Three categories. One root cause. The action executed and nothing stopped it.
Read articleFive Eyes Governments Establish Human Oversight of Agentic AI is Not Optional.
Six national cybersecurity agencies published joint guidance in May 2026 declaring that human oversight of agentic AI is not an optional safeguard but an essential prerequisite. The requirement has been named. The product that implements it has not.
Read articleThe UK Government Is Right to Lock the Door. The Vault Needs a Different Lock.
Cyber Essentials is necessary and worth having. It does not address what happens when the door is opened legitimately, which is how most serious incidents actually occur.
Read articleYou Can Change Your SSVC Scores. Here Is What That Means in Practice.
CISA's vulnerability prioritisation framework classifies the most urgent vulnerabilities as Act. GoFirm formally converts a category of Act vulnerabilities into Track or Attend. That is not a marketing claim. It is a logical consequence of applying CISA's own methodology to an environment where the execution boundary is protected.
Read articleNIST Just Proved AI Guardrails Cannot Hold. So Don’t Rely on Them.
A mathematical proof published by NIST shows that no finite set of AI guardrails can be universally robust against adversarial prompts. The industry’s answer is a permanent, expensive red-teaming programme. GoFirm offers a different calculation.
Read articleBlog content generation is AI assisted.