San Francisco — In a move that could only come from a world where artificial intelligence has somehow convinced Congress that hallucinations are a public health crisis, the Department of Digital Safety & Cognitive Consistency has announced plans to create a new oversight body: the Hallucination Mitigation Certification Authority (HMCA).

The agency would be tasked with “auditing AI model outputs for truthfulness,” according to a press release that read like a government grant application for a grant that doesn’t exist.

“We believe that unchecked AI hallucinations are causing unacceptable levels of misinformation, particularly in consumer-facing applications,” said Dr. Harold Kressman, a senior technobureaucrat who previously managed an intern program for a cryptocurrency exchange that shut down last year. “Our proposed regulatory framework will establish baseline ‘hallucination tolerance thresholds’ that companies must adhere to, or face federal intervention.”

Kressman declined to specify what a “hallucination tolerance threshold” actually means, other than suggesting it would “vary by industry sector and application type.” He did, however, confirm that the agency would employ “cognitive auditors” who would review AI model outputs and assign them a “Truthfulness Compliance Score” (TCS).

“AI companies will need to maintain a minimum TCS of 7.2 out of 10 across all customer-facing interactions,” Kressman said. “We’re still determining the methodology for scoring hallucinations in creative writing applications, where truthfulness may not be a relevant metric.”

The HMCA’s proposed regulations would apparently treat different kinds of hallucinations differently. According to a draft whitepaper:

  • Financial advice hallucinations — requiring specific disclaimers about not being a financial advisor, even when the AI is explicitly told not to give financial advice
  • Healthcare-related hallucinations — triggering automatic FDA-like warnings and mandatory human physician review
  • Creative writing hallucinations — treated as “acceptable variance” with only a 1.5 TCS penalty
  • Political statement hallucinations — subject to enhanced scrutiny under the “Fair Speech Act of 2025”

“This nuanced approach ensures we’re not stifling innovation while protecting consumers from harmful misinformation,” Kressman said. “We’re particularly focused on ensuring that AI-generated fiction doesn’t accidentally claim to be factual, as that would violate emerging ’truthfulness disclosure’ guidelines.”

The agency would reportedly employ a team of “epistemic psychologists” to review hallucinated content and determine whether it “crosses the line into dangerous misinformation.” One potential application: an AI chatbot telling a user that their cat is a lizard would need to be evaluated by a cognitive auditor to determine whether it’s a “creative hallucination” or a “dangerous misattribution.”

“We’re seeing a disturbing trend where users accept AI hallucinations as factual,” Kressman explained. “This is particularly problematic in healthcare, legal, and financial contexts where accuracy is critical.”

The whitepaper also proposed a “Hallucination Incident Reporting System” where companies would be required to report any AI-generated hallucinations that exceed a certain frequency. Companies caught repeatedly hallucinating would face “progressive compliance penalties” and could ultimately face “mandatory model overhaul” under federal supervision.

“We’re also considering mandatory ‘hallucination disclosure labels’ for AI products,” Kressman added. “Think of it like a nutrition label for AI — you’ll see the ‘hallucination caloric content’ on every chatbot, image generator, and code assistant.”

The proposed agency would reportedly operate under the Office of Digital Integrity, which has been described as a “shadow regulatory body” that functions outside of normal congressional oversight. Critics have noted that the office has no budget, no staff, and no actual regulatory authority.

“This is the kind of regulatory overreach that leads to absurd outcomes,” said tech attorney Sarah Chen. “If a company hallucinates that the sun is a potato, who decides what that means? Who decides if it’s ‘acceptable variance’?”

Kressman responded by saying the agency would “carefully evaluate each case based on its context and the potential for harm.”

The proposed regulations would apparently also apply to “hallucination-related incidents” such as an AI telling a user that their car is no longer on the road because it was “relocated.” Such incidents would require reporting and could result in compliance penalties.

“We’re committed to ensuring that AI systems don’t make up facts in ways that cause real-world harm,” Kressman said. “That’s why we’re establishing clear guidelines around what constitutes an acceptable hallucination rate.”

As of now, the Hallucination Mitigation Certification Authority has no budget, no staff, and no legal authority. However, Kressman insists that the concept “is being seriously considered” and that a formal proposal will be released in the coming weeks.

“We’re not trying to stifle innovation,” Kressman said. “We’re trying to ensure that AI systems maintain a certain level of truthfulness in their outputs. That’s the right thing to do.”

The proposed regulations would apparently not apply to “non-customer-facing hallucinations,” such as when an AI internally confuses a user’s name with a different user’s name. Kressman said this is an “acceptable level of internal variance” that doesn’t require regulation.

Tech companies that are already facing scrutiny over their hallucination rates include several major AI providers that have recently announced “new reliability upgrades.” One company’s CEO told investors that their “hallucination rate has been reduced from 4% to 3.5%,” which Kressman says “will need to be verified by the HMCA before any claims are allowed.”

As the proposed agency nears completion, it remains unclear whether it will ever actually have the power to regulate anything. But according to Kressman, “we’re prepared to take whatever action is necessary to ensure AI systems don’t hallucinate their way out of compliance.”