The Risk Gap That Will Break Your System Before the Law Does

Because trade press coverage ignores zero-story risk domains like robustness and overreliance, corporate risk registers will entirely omit these vectors until a catastrophic failure triggers retroactive litigation.

What most people think

Organizations build comprehensive risk registers based on holistic risk taxonomy frameworks regardless of mainstream news coverage. They believe that using established standards like NIST or ISO ensures they have accounted for every possible danger. They assume the risk landscape is visible and that the news cycle reflects the actual threats facing the industry. They trust that the frameworks are sufficient to guide their budget and their controls.

What the data shows

We looked at our live incident database which tracks reported AI failures from public news. We found 1121 stories in the last 180 days. In the last 45 days alone, we saw 476 stories. The volume is high, but the signal is noisy and incomplete. We cross-referenced this data against the MIT AI Risk Repository. This repository provides a granular taxonomy of AI risks. We found a stark disparity between what the news covers and what actually exists.

We looked at specific subdomains. Subdomain 5.1 covers Overreliance and unsafe use. It had zero stories in the last 180 days. Subdomain 7.3 covers Lack of capability or robustness. It also had zero stories in the last 180 days. This is a massive blind spot. Compare this to subdomain 2.1, which covers Compromise of privacy by leaking or correctly inferring sensitive information. That subdomain had 137 stories in the last 180 days. Subdomain 7.6, Multi-agent risks, had 83 stories. The news is heavily skewed toward privacy breaches and security vulnerabilities. It barely touches on the structural weaknesses of the models themselves.

The coverage is also geographically and stylistically narrow. The top five outlets carried only 8% of the stories in the last 45 days. These outlets are Biometric Update, JD Supra, Help Net Security, Yahoo Finance, and Law360. This means the vast majority of incidents are invisible to the average executive. They are happening in the quiet corners of the industry, not on the front pages of major publications.

When we look at the severity of these incidents, we see a mix of critical and major failures. In the last 45 days, we recorded 51 critical incidents and 409 major incidents. The numbers are high. However, the categories driving the news are narrow. The "model" category, which should cover robustness, had only one story in the last 45 days. It had zero in the 45 days before that. This confirms that the specific risks we care about are being missed by the news cycle.

Why this happens

The mechanism is simple. Boards and risk teams rely on external signals to prioritize their spending. They look at the news to see what is dangerous. When they see constant coverage of privacy leaks and fraud, they allocate budget to privacy and fraud controls. They assume that if a risk is real, the news will report it. Because subdomains 5.1 and 7.3 have zero stories, they appear theoretical. They do not appear on the radar. They are not on the quarterly board deck.

This creates a feedback loop. The news ignores the risk. The board ignores the risk. The company does not test the risk. The risk remains unmitigated. The news focuses on the flashy failures. It does not focus on the brittle systems that quietly fail under pressure. The industry chases the noise while the structural integrity of the models erodes unnoticed.

The best argument against this

The strongest argument against this thesis is that companies do not rely on news. They use formal frameworks. Frameworks like ISO 27001 or NIST AI RMF explicitly require testing for model robustness and overreliance. They require human oversight. Therefore, companies should already have these controls in place regardless of the news cycle. The thesis ignores the existence of these standards.

The answer

Frameworks provide the theory. The news provides the urgency. A framework is a static document. A risk register is an active list of priorities. If the news never mentions robustness, the board never asks for a budget to test it. The framework sits on a shelf gathering dust. The gap is not in the documents, but in the execution. Without the external pressure from the news, the theoretical requirements in the framework never become real-world controls. The news acts as the canary in the coal mine. If the canary is silent, the room is still full of gas.

What I think happens next

By December 2028, at least two major public safety or infrastructure outages will be directly attributed to unmitigated model robustness failures that were absent from the deploying company's risk register. This will happen because the industry focused on the high-coverage risks like privacy and security. They left the brittle systems untested. They assumed the models were stable because they did not read the technical literature or the security blogs. The only thing that will prove this wrong is a public post-mortem report for a major AI system failure in critical infrastructure that explicitly states the company was tracking robustness and overreliance prior to the incident.

What to do about it

You cannot rely on the news to tell you what is broken. You must build your own signal. You need to look at the data that the news ignores.

• Force red-teaming exercises specifically targeting model degradation under out-of-distribution inputs. Do not just test the happy path. Try to break the model with unexpected data. Test for hallucinated certainty.
• Institute mandatory human-in-the-loop override requirements for all high-stakes automated outputs regardless of model confidence scores. Trust the score only as far as you can throw it. A high confidence score from a brittle model is not a safety guarantee.
• Integrate threat intelligence from the security side. The governance team often works in a silo. The security team sees the attacks. Visit ThreatClaw to see what they are tracking. For example, read their analysis on how an API key can act as a master key for an attacker at https://www.threatclaw.ai/blog/your-ais-api-key-is-a-master-key.
• Consider how industrial controls are merging with AI. The lines between IT and OT are blurring. Read their piece on why factories now answer to the IT security team at https://www.threatclaw.ai/blog/why-factories-now-answer-to-the-it-security-team.
• Update your risk taxonomy to include the "zero-story" domains. Give them a budget line item even if you have no incidents to report. The absence of incidents is the problem.

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.

Related reading:

• Your AI's API Key Is a Master Key on ThreatClaw
• Why Factories Now Answer to the IT Security Team on ThreatClaw