When A.I. Floods the System

A New Kind of Administrative Strain

From community colleges in the United States to employment tribunals in Britain, public institutions are confronting a similar problem: generative artificial intelligence is making it cheaper and easier to flood systems with participation that looks real enough to pass initial scrutiny.

In one case, scammers have used fake or stolen identities to enroll “ghost students” in online college courses, sometimes relying on bots or AI tools to submit assignments long enough to unlock financial aid payments. In the other, British judges say AI-assisted filings are contributing to a wave of longer, more complicated employment claims, some padded with invented legal citations, that are slowing justice for workers with legitimate grievances.

The common thread is not simply deception. It is overload.

Institutions built on the assumption that filing a claim, enrolling in a class or completing coursework requires meaningful human effort are now being tested by software that can generate essays, forms and legal arguments in seconds. What once took time, knowledge or organization can now be done at scale and at little cost.

Colleges Battle “Ghost Students”

Community colleges, especially those with large online offerings, have become a prime target for student-aid fraud. Criminals enroll fictitious students, or use stolen identities, in order to secure federal or state financial aid. To keep those fake students from being dropped early in the term, they may use generative AI to complete introductory assignments and discussion posts, creating the appearance of legitimate attendance.

Faculty members have described classrooms filled with students who never meaningfully engage but continue to submit passable, machine-produced work. The effect can be more than academic. Real students may find themselves shut out of oversubscribed classes, while colleges can be left to sort through the administrative and financial fallout.

The scale of the problem has been particularly visible in California’s community-college system, where officials in 2021 suspected that roughly one in five applicants was fraudulent amid a boom in online learning. Fraud surged in the years that followed, peaking in early 2025 before beginning to ease. State officials say losses have declined, but the schemes persisted into 2026.

At the federal level, the Education Department says it has prevented more than $1 billion in student-aid fraud since January 2025 and has expanded screening of FAFSA submissions nationwide. California has also tightened protections, with mandatory ID verification taking effect on July 1, 2026.

Those safeguards may reduce losses, but they also underscore a difficult trade-off. Every new check meant to stop fraud can create new hurdles for legitimate students, particularly low-income applicants, first-generation students and others who already face bureaucratic barriers to enrollment.

British Tribunals Face a Paper Avalanche

A similar dynamic is playing out in Britain’s employment courts, where judges and administrators are grappling with a steep rise in claims and a mounting backlog.

Employment tribunals received 50,000 single claims in 2025-26, and 64,000 single claims were still outstanding at the end of March 2026, a 55 percent increase from a year earlier. Judiciary minutes have identified AI as at least a partial cause of claims becoming more complex and time-consuming to process.

Judges have reported seeing filings generated with tools such as ChatGPT and Grok that run to hundreds of pages, burdened by repetition, weak reasoning or fabricated authorities. The result is not merely more paperwork, but more judicial time spent separating genuine arguments from automated filler.

That burden falls hardest on people the system is designed to help. Workers bringing legitimate complaints over wages, discrimination or unfair dismissal can wait longer as tribunals sift through unreliable or inflated submissions.

Britain’s judiciary has already begun responding. Updated guidance issued in October 2025 warned judges and court users about AI “hallucinations,” confidentiality risks and the danger of relying on generated legal material without verification. A working group of the Civil Justice Council is now considering whether additional rules are needed for AI-assisted document preparation, especially in witness statements and cases involving people without lawyers.

Still, the precise role of AI in the surge remains uncertain. Labor-market conditions, changes in employment practices and other legal factors may also be driving claims. What judges appear increasingly confident about is that AI is changing the character of what arrives at the courthouse door: more text, more apparent polish and, often, more hidden unreliability.

Systems Built for Scarcity, Confronting Abundance

These episodes point to a broader governance challenge of the AI era. Generative tools do not need to be perfect to be disruptive. They only need to be good enough to pass the first gate.

A college need not be convinced that a student is brilliant; it need only fail to recognize quickly that the student does not exist. A tribunal need not accept a legal argument as sound; it need only spend time untangling whether it is real. In both cases, the cost of participation has fallen dramatically for bad actors, while the cost of verification remains stubbornly human.

That mismatch is becoming a defining administrative problem. Public systems often depend on triage, trust and limited staffing. When AI enables a flood of plausible applications, assignments or claims, the burden shifts to teachers, clerks, judges and financial-aid offices that must verify authenticity one case at a time.

The consequences are practical and immediate: delayed aid, crowded classes, repayment disputes, clogged dockets and longer waits for redress. But they are also structural. If institutions respond with stricter screening, they risk excluding legitimate users. If they do not, they risk being overwhelmed.

What Comes Next

Neither the United States nor Britain yet has a full accounting of the damage. Federal agencies have not published a comprehensive national estimate of ghost-student losses, and British officials have not quantified how much of the tribunal backlog is directly attributable to AI-generated filings.

But the pattern is becoming harder to dismiss as isolated misuse. Across sectors, generative AI is exposing a vulnerability in systems designed for an older information environment, one in which producing a convincing essay, application or legal pleading required enough time and effort to act as a natural brake.

That brake is weakening. And as institutions race to adapt, the central question is no longer whether AI will be used to game public systems. It is whether those systems can be redesigned quickly enough to protect access for the people they were meant to serve.

Sources

Further reading and reporting used to add context:

AI News