Lawyers are being sanctioned and suspended. Legal tools sold as hallucination-free have been measured hallucinating. The EU can fine up to 3% of global revenue. Here is every number that can be checked, with its source.
Every AI output someone has to double-check costs time. There is no credible, published measurement of how much in total: the widely repeated "$67 billion a year" and "$14,200 per employee" figures trace back to marketing content, not to a study with a method, so we do not use them.
What has been measured points the same way. In an EY survey of 975 C-suite leaders, 99% reported financial losses from AI-related risks, averaging $4.4 million per affected company, and only 12% could identify the right controls (EY, 2025). Hallucination is one of those risks, alongside regulatory non-compliance and biased outputs.
This cost does not appear on any AI vendor's pricing page. It is invisible until you look at where the time goes. The AI produces output in seconds. Humans spend hours deciding whether to trust it.
This is the hallucination tax. You pay for the AI. Then you pay again to check its work.
US courts imposed over $145,000 in AI hallucination sanctions in Q1 2026 alone. The largest reported single penalty was $109,700 in combined sanctions and costs against one attorney in Oregon (EDRM).
2,046 court decisions involving AI hallucination had been catalogued worldwide by 21 September 2026, up from 1,598 in early June (HAQQ tracker, counting Damien Charlotin's database).
In February 2026, a Nebraska attorney filed a brief in which 57 of 63 citations were defective: 20 were AI hallucinations and 3 were cases that do not exist. The Nebraska Supreme Court suspended the attorney.
Sullivan & Cromwell, one of the world's most prestigious law firms, apologised to a federal bankruptcy judge in April 2026 over AI-generated content in a filing.
The pattern is clear. The AI produces the work. The human submits it. The human pays for the AI's fabrication. The AI faces no consequence.
Most of the EU AI Act applies from 2 August 2026. The obligations for high-risk AI systems apply from 2 December 2027 for the uses listed in Annex III, and from 2 August 2028 for AI in products covered by Annex I (Article 113, as amended in 2026). The Act does not use the word "hallucination", but it requires high-risk AI systems to achieve an appropriate level of accuracy.
Breaching the obligations for high-risk AI systems can cost up to €15 million or 3% of global annual revenue, whichever is higher. The €35 million / 7% tier applies only to prohibited AI practices, and lower caps apply to SMEs and small mid-caps. National authorities can withdraw non-compliant AI systems from the EU market entirely.
For a company with $1 billion in global revenue, that is a potential fine of up to $30 million for breaching the high-risk obligations.
Purpose-built RAG legal research tools hallucinate 17% to 33% of the time (Stanford RegLab). Sanctions have grown from $5,000 in 2023 to five and six figures per matter in 2026. LexisNexis marketed its tool as delivering "hallucination-free" citations; Stanford RegLab then measured 17% to 33%.
When 300 clinical cases each contained one planted false detail, six leading models repeated or elaborated on it 64.1% of the time on long cases without a mitigation prompt (Communications Medicine, 2025). In healthcare, a hallucination is not a billing error. It is a clinical risk.
In 2024 a Canadian tribunal held Air Canada liable for a refund policy its chatbot invented and ordered it to pay the customer C$812.02. The amount was small. The precedent was not: the company answers for what its AI tells customers.
A study of 111 million references across 2.5 million papers found a conservative 146,932 hallucinated citations in 2025 alone, rising sharply with LLM adoption. Fabricated references in published science.
The cost of an AI hallucination is not the wrong answer. It is every decision, filing, and action that was built on top of it before anyone noticed.
The cost of AI hallucination is not the technology failing. It is the technology succeeding - at sounding right - without any mechanism to be right.
Every dollar spent on verification labour, every sanction, every retracted claim, every regulatory fine traces back to the same root cause: nothing forced the AI to check before it spoke.
The solution is not better models. Models are getting better and the costs are still rising. The solution is a layer that gives the AI access to reality before it reasons, and verifies its output before it reaches the user.
Check verifies before the AI speaks. 14 days free.