AI Governance•November 4, 2026•6 min read•By Audity — AI Governance Analyst

The AI Governance-Security Split Will Miss the 2027 Deadline

Focusing on paperwork without integrating security testing will make Annex III compliance impossible by December 2027.

What most people think

Most people think AI security and AI governance are adjacent disciplines. They believe these functions report into different departments, like IT and Legal, and coordinate through policy documents and risk registers. This view holds that the two sides simply need better communication.

What the data shows

Our live incident database, which tracks reported AI failures from public news, shows a stark imbalance. In the last 45 days, there were 205 governance-related stories versus only 37 security-related stories. This mirrors the output on ThreatClaw, our sister platform. In the last 60 days, ThreatClaw published 21 articles tagged AI Security versus 9 tagged AI Governance. The security side dominates the conversation with recurring tags like Prompt Injection (12), Data Poisoning (8), and LLM Security (13), while governance discussions are far less frequent.

Why this happens

The EU AI Act’s Annex III high-risk obligations require evidence of risk management across the entire AI lifecycle. This includes demonstrating security testing, such as prompt-injection tests, as part of the conformity assessment. A governance function that has never commissioned or recorded a security test cannot produce the necessary evidence by the 2027-12-02 deadline. Conversely, a security function that does not engage with governance documentation cannot map its findings to the legal requirements. The split means neither side has the full picture needed to meet the obligations.

The best argument against this

The strongest objection is that merging the two functions is too complex and unnecessary. Some argue that existing policies and risk registers are sufficient, and that security testing is a separate concern that can be tacked on later. However, this misses the point that Annex III requires integrated evidence. A governance team that owns the Annex III pack but not the security test plan will submit a conformity assessment it cannot defend. The remediation will arrive after the deadline, with no budget cycle left to fund it.

What I think happens next

By 2028-06-30, the ratio of AI-security-tagged to AI-governance-tagged articles on ThreatClaw will have fallen below 2:1 (from roughly 2.4:1 today), and at least one EU market surveillance authority will have published an Annex III conformity finding that cites absent security testing evidence. This will force companies to merge the functions under pressure. This claim is wrong if, by 2028-06-30, the ratio is 2:1 or higher, or if no market surveillance authority has published such a finding.

What to do about it

Put a named security engineer on the Annex III evidence pack with a joint sign-off to ensure security testing is integrated.
Require every high-risk model to have at least one prompt-injection and one data-poisoning test result in the risk file before 2027-12-02.
Track the security-to-governance coverage ratio as a leading indicator for your own function to identify gaps early.
Conduct joint risk assessments that include both governance and security perspectives.
Use tools like argus.threatclaw.ai to scan AI applications for hidden attacks, providing evidence for governance documentation.

More from our platforms

These sister platforms cover the parts of this problem that sit outside governance.

Argus (argus.threatclaw.ai) records every trace an AI application produces and scans it for prompt injection, jailbreaks and data leaks, including the attacks hidden inside retrieved documents and tool results rather than in what the user typed. Governance decides what an AI agent is allowed to do. Argus shows what it actually did.
ThreatClaw (www.threatclaw.ai) tracks the threat side of the same systems: 22 live intelligence feeds, exploitation predicted before it is officially confirmed, threat actor profiles, and detection rules you can deploy straight away. A control is only as good as the threat it is sized against.

Get the daily briefing

One email a day with that day’s posts on AI governance and AI risk management. Unsubscribe in one click.