Colombian Court Rejects Appeal for Alleged AI Writing-Then Its Own Ruling Is Flagged as AI-Generated
Colombia’s highest criminal court has found itself at the center of an AI-era paradox. The Supreme Court rejected a legal appeal after concluding the filing was written with the help of artificial intelligence-only for the court’s own ruling in the same case to be flagged as likely AI-generated by the very type of detector it relied on.
The episode has ignited a debate about fairness, legal ethics, and whether AI-detection tools are reliable enough to influence outcomes in courtrooms.
Court Throws Out Appeal Over Suspected AI Use
The case concerned a cassation appeal filed before Colombia’s Supreme Court of Justice, the top authority in criminal matters. The court refused to even analyze the substantive arguments of the filing, arguing that the brief had not been drafted by a human lawyer but produced-at least in large part-by an AI system.
To support that conclusion, the justices relied on the output of an AI-detection program. According to the court, the text exhibited a pattern and probability profile closely associated with generative AI models, prompting what it called a “well-founded suspicion” that the attorney had unlawfully delegated their work to a machine.
Instead of treating AI as a neutral tool akin to a spell-checker or research assistant, the court framed its use as a potential violation of professional duties. It emphasized that legal documents submitted to the Supreme Court must be the result of the attorney’s own intellectual effort and judgment, not the product of automated systems.
The Same Detector Turns on the Court’s Own Text
The story took a striking turn when the attorney whose filing was rejected decided to scrutinize the court’s reasoning using the same category of AI-detection software. When the lawyer ran the court’s own written decision through the detector, the tool reportedly concluded-with a confidence score of 93%-that the ruling itself was likely generated or heavily assisted by AI.
In other words, the very kind of evidence used to disqualify the lawyer’s work could be used, under the same logic, to cast doubt on the court’s own authorship. The result highlights a deep inconsistency: if an AI score is enough to disqualify a filing from a lawyer, what should it mean when the same type of tool flags a judicial ruling?
The court has not publicly acknowledged drafting its opinion with AI tools, and there is no conclusive proof that it actually did. But the episode underscores a crucial point: AI-detection results, especially when treated as definitive, can easily produce contradictory or absurd outcomes.
Double Standard or Broken Technology?
This clash between the court and its own “evidence” raises two uncomfortable possibilities.
One is the appearance of a double standard. If a high AI-probability score is sufficient to discredit a lawyer’s document, it seems inconsistent to ignore a similarly high score when it implicates the court’s own work. Even if the judges insist their ruling was written by humans, the tools they relied on appear unable to distinguish that from machine-generated text.
The other possibility is more straightforward: the detection technology itself may be unreliable or misunderstood. AI classifiers are probabilistic, not absolute. They analyze patterns such as word choice, sentence structure, and statistical regularities, then generate a likelihood that text came from an AI system. Those probabilities can be wrong-sometimes dramatically so-especially when assessing formal or highly structured writing, like legal opinions.
In practice, the tools can end up labeling careful, clear human writing as AI-like simply because it resembles the style that large language models also tend to produce.
What AI Detectors Can and Cannot Do
Independent examinations of AI-detection tools have already found that:
– They frequently produce false positives, labeling human-written work as AI-generated.
– Their performance can degrade over time as AI models evolve and mimic human style more closely.
– They can be biased against certain writing styles, levels of fluency, or non-native speakers whose patterns differ from the data the detectors were trained on.
– Minor edits to text-such as paraphrasing or adding small changes-can drastically change the score, revealing how fragile these assessments are.
Within this context, treating detection scores as hard proof rather than as noisy indicators becomes especially problematic. In academic institutions, media outlets, and workplaces, experts increasingly warn against using such tools as the sole basis for punishment or sanctions. The Colombian court’s situation illustrates why that caution is necessary in the legal arena too.
The Legal Ethics Question: Is AI Assistance a Violation?
Beyond the technology, the case exposes an unresolved ethical question for the legal profession: when, if ever, is it acceptable for lawyers to use AI tools?
Most legal ethics frameworks agree on at least three core duties: competence, confidentiality, and honesty. AI intersects with each:
– Competence: Lawyers must understand the tools they use well enough to avoid errors, hallucinations, or misinterpretations. Blindly copying AI-generated text without verification would likely breach this duty.
– Confidentiality: Feeding sensitive client information into public AI systems could expose private data and violate professional secrecy.
– Honesty and independence: Courts expect that filings represent the lawyer’s own analysis and legal reasoning, not unverified content taken from a machine.
Yet using AI does not automatically mean a lawyer has violated these principles. Many attorneys rely on software for research, drafting templates, translation, or formatting. When AI is used as an assistant under the lawyer’s supervision, with the lawyer ultimately responsible for every word, that can be compatible with ethical standards.
The difficulty lies in drawing a clear line between acceptable assistance and impermissible delegation-and deciding how, if at all, that line should be enforced.
Should AI Detectors Have Any Place in Courtrooms?
The Colombian decision illustrates the risks of importing AI-detection tools into the justice system without a clear framework. There are several dangers:
1. Erosion of due process: If a court dismisses filings based largely on detection software, litigants may be punished without solid evidence of wrongdoing.
2. Lack of transparency: Many detection tools function as black boxes. Lawyers and parties cannot meaningfully challenge or understand why a text was flagged.
3. Inconsistent application: As this case shows, the same type of tool can be ignored when it produces inconvenient results, such as suggesting a ruling is AI-assisted.
4. Chilling effect: Lawyers may avoid beneficial technologies out of fear that any trace of AI-style phrasing could undermine their filings.
A more balanced approach would treat AI-detection outputs, if used at all, as preliminary indicators prompting further inquiry-not as conclusive proof. Courts could, for example, invite explanations from counsel instead of reflexively rejecting filings.
How Courts Could Handle AI Use More Fairly
Rather than banning AI by implication or relying heavily on unreliable detectors, courts could adopt explicit, balanced rules. Such a framework might include:
– Disclosure duties: Require lawyers to disclose significant AI assistance in preparing filings, especially if AI contributed to legal reasoning or drafting large portions of text.
– Human accountability: Make it clear that regardless of tools used, the attorney remains fully responsible for accuracy, originality, and ethical compliance.
– Confidentiality safeguards: Prohibit feeding privileged or sensitive case information into public AI systems without proper protections.
– Sanctions based on conduct, not tools: Discipline should target actual misconduct-such as plagiarism, false statements, or breach of confidentiality-not mere use of AI itself.
– Expert review instead of automated scores: When concerns arise, an independent human expert could review a document’s originality and quality, rather than relying on probabilistic software.
Such measures would protect the integrity of proceedings while recognizing that AI is likely to become a routine part of professional work, including law.
The Broader Implications for Trust in Justice
At its core, the Colombian episode is not just about software or technical accuracy; it is about trust. Courts derive their authority from the perception that they act consistently, rationally, and transparently. When a court leans on an opaque algorithm to invalidate a filing but brushes aside similar results when its own work is tested, confidence in impartial justice can erode.
If detection tools are unreliable enough to misclassify a judicial ruling as AI-generated, then basing life-altering decisions on those tools becomes difficult to justify. The incident serves as a warning: technological shortcuts in assessing credibility or authorship can backfire, particularly in environments where accuracy and fairness are paramount.
Why This Case Matters Beyond Colombia
Although this controversy unfolded in Colombia, its implications stretch far beyond one country’s borders. Around the world, courts, universities, and employers are scrambling to respond to the rise of generative AI. Many are tempted to turn to AI-detectors as easy gatekeepers.
However, as the Colombian ruling and its aftermath show, these tools can become traps. They can:
– Penalize innocent users whose human writing happens to resemble AI style.
– Produce contradictory outcomes that undermine institutional credibility.
– Create incentives to disguise or “launder” AI-assisted work rather than use it responsibly and transparently.
The key lesson is not that AI should be banished from legal practice or public institutions, but that simple technological fixes cannot replace thoughtful regulation, ethical reflection, and human judgment.
Toward Clearer Rules for an AI-Driven Legal Future
As generative AI becomes more powerful and ubiquitous, courts will increasingly confront filings, evidence, expert reports, and even witness statements that may involve AI at some stage. Ignoring that reality is unrealistic. Overreacting with blanket suspicion is equally dangerous.
The Colombian Supreme Court’s decision-and the ironic flagging of its own ruling-may ultimately serve as a catalyst for clearer policies. Legal systems will need to:
– Define when AI use is compatible with professional duties.
– Specify what kinds of disclosure are required.
– Decide whether AI-detection tools have any legitimate role, and if so, strictly limit how their results can be used.
– Invest in training judges and lawyers to understand both the capabilities and limits of AI.
Until then, cases like this will continue to expose the tension between rapidly evolving technology and institutions built on long-standing principles of fairness, evidence, and accountability.
In that sense, the Colombian court did more than reject a single appeal. It inadvertently demonstrated that without robust rules and cautious use of technology, the justice system itself can be tripped up by the very AI tools it turns to for help.