The 2027 Gamble: Why Firms Will Downgrade AI Incidents to Avoid Scrutiny

Firms will game the EU AI Act's serious-incident reporting rule by downgrading incident severity to avoid scrutiny, mirroring the sharp drop in critical incidents we are already seeing in public data.

What most people think

Most people believe that regulators will catch under-reporting through audits and whistleblowers, so firms will comply with the letter of serious-incident rules. The prevailing view is that the threat of legal penalties and reputational damage will force companies to be honest about their failures. People assume that the transparency requirements of the new framework will create a level playing field where bad actors cannot hide their mistakes from the public eye.

What the data shows

Our live incident database, which tracks reported AI failures from public news, reveals a disturbing trend that contradicts the idea that risks are actually decreasing. In the last 45 days, critical incidents fell by nearly 50 percent, dropping from 94 stories to just 48. Meanwhile, major incidents rose slightly from 370 to 381, and minor incidents fell from 12 to 8. This shift suggests that the definition of "critical" is becoming looser, even though the actual number of severe failures in the news has not decreased.

The data also shows a massive drop in governance-related reporting, with stories in that category falling by 51. This decline indicates that companies are becoming more secretive about how they manage AI risks internally. We see a similar skew in the types of incidents reported. Categories like "overreliance and unsafe use" and "environmental harm" have generated zero stories in our 180-day window, while categories like "privacy" and "security vulnerabilities" dominate the headlines. This disparity means the public data is incomplete and likely skewed by what is easiest to report rather than what is most dangerous.

Why this happens

The mechanism driving this behavior is simple economics. Reporting a serious incident triggers legal scrutiny, fines, and potential reputational damage. The EU AI Act's serious-incident reporting obligation is not due until December 2027, giving firms a clear window to manipulate their internal classifications. In the meantime, there is no public audit mechanism specified to verify a company's classification of an incident as minor or major. Downgrading an incident from critical to major or minor is a low-cost way for firms to avoid the immediate costs of compliance and regulatory pressure.

The pressure is compounded by the upcoming deadlines for other regulations. With the Colorado ADMT Act and California CPPA requirements taking effect in early 2027, companies are already rushing to establish basic compliance frameworks. They will view incident reporting not as a tool for safety, but as a compliance cost to be minimized. By classifying incidents as minor, they can claim compliance with the new rules while avoiding the deep dive that comes with a serious incident designation.

The best argument against this

The strongest objection is that regulators will catch this behavior during mandatory audits. If a regulator digs into a company's internal logs and finds a pattern of downgrading or hiding incidents, they will impose heavy fines. The logic is that the legal liability of getting caught hiding a serious incident will far outweigh the savings gained from not reporting it.

However, the early phase of the EU AI Act lacks the resources to audit every high-risk system independently. Without a requirement for third-party verification of incident classifications, companies can easily manipulate their own internal reporting standards without fear of immediate detection. Audits are resource-intensive and typically target the most egregious offenders. A company can systematically downgrade incidents for years, building a paper trail of minor classifications that looks compliant on the surface, while the actual risk grows beneath the radar.

What I think happens next

By June 2028, a named EU regulator or an industry survey will publicly report that over 30 percent of AI incidents classified as "minor" by firms were later reclassified as "serious" upon review. This would prove that the voluntary nature of the reporting system has failed and that companies are systematically under-reporting to avoid regulatory pressure. The gap between internal classifications and actual severity will become too large to ignore, forcing a crisis of confidence in the reporting data.

What to do about it

• Build an internal severity-classification rubric with independent audit trails before the December 2027 deadline.
• Prepare a legal defense strategy now for why you classified a specific incident as minor, so you do not panic when a regulator asks.
• Lobby for regulator-published anonymized reclassification statistics to level the playing field.
• Start tracking the gap between your internal incident counts and public news reports to spot potential under-reporting in your own organization.
• Conduct a stress test of your incident reporting procedures to ensure they can withstand external scrutiny.

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:

• When Your AI Fails, Who Do You Tell? The New Rules Are Here on ThreatClaw
• ThreatClaw Intelligence Brief — 2026-08-24 on ThreatClaw