Attorneys are citing to cases which don't exist

It's not only becoming common for law firms to submit briefs to courts which they drafted with the assistance of artificial intelligence software, courts are catching them in the act and finding that some of the caselaw cited to in these briefs is completely fictional. AI 'hallucinations' are instances where AI software generates fallacious information in response to a question.

In January, the Second Circuit issued a per curiam decision, Op., Park v. Kim, No. 22-2057 (2d Cir. Jan. 30, 2024), ECF No. 178-1, in which the court found that:

The attorney who filed the brief was referred to the Court's Grievance Panel. The Court cited the 5th Circuit's amendment to its rules which requires that attorneys certify that no generative artificial intelligence was used for a filing, or that at least that the filing was reviewed for accuracy by a human.

In May 2023, Judge Castel of the United States District Court for the Southern District of New York, issued an Order to Show Cause why the plaintiff's counsel should not be sanctioned pursuant to Fed. R. Civ. P. 11 and the Court's inherent authority for submitting an affirmation in opposition to a motion to dismiss which included citations to six "bogus judicial decisions with bogus quotes and bogus internal citations." Order to Show Cause at 1, Mata v. Avianca, Inc., No. 22-cv-1461 (PKC) (S.D.N.Y. May 4, 2023), ECF No. 31. The brief used citations to reporters which actually refer to other cases:

Id. at 2.

Not only did the plaintiff's affirmation refer to Varghese v China South Airlines Ltd., 925 F.3d 1339 (11th Cir. 2019), and other cases which are completely made up, but when the court issued an order asking that plaintiff's counsel submit copies of these cases (Order, Mata v. Avianca, Inc., No. 22-cv-1461 (PKC) (S.D.N.Y. Apr. 11, 2023), ECF No. 25) the counsel in turn filed an affidavit submitting AI generated copies of the imaginary cases! Affidavit, Mata v. Avianca, Inc., No. 22-cv-1461 (PKC) (S.D.N.Y. Apr. 25, 2023), ECF No. 29.

A copy of Varghese was filed on PACER, but it's something that AI simply invented:

PACER doesn't lie!

At a subsequent hearing, Judge Castel excoriated the plaintiff's attorney for submitting the fictitious case:

Hr'g Tr. at 15:17-17:6, Mata v. Avianca, Inc., No. 22-cv-1461 (PKC) (S.D.N.Y. Apr. 11, 2023), ECF No. 52. So the case that ChatGPT invented was not even one which had an internal logic of its own.

The court sanctioned the plaintiff's attorney under Rule 11, fined him $5,000 and ordered him to send a letter to each judge listed as the author of the false cases the affirmation cited to. Mata has been dismissed for being untimely under an international convention that covered a claim for an injury suffered by the plaintiff during an international flight.

EU 2024/1689 - Europe Tries to Reign in AI

This week the EU regulation 2024/1689, "laying down harmonised rules on artificial intelligence" became effective. The European Artificial Intelligence Act will regulate the use of artificial intelligence with an aim towards protecting the rights and the safety of EU citizens.

The regulation does not seek to restrict spam or AI that suggests products to consumers. It will require chatbots to disclose to people who they are communicating with that they are in fact AI and not a human being. Generative AI, such images or video created by AI, will need to be flagged as content created by artificial intelligence.

Paragraph 30 of EU regulation 2024/1689 forbids using biometric data to predict a person's sexual orientation, religion, race, sexual behavior, or political opinions, although it provides for an exception for filtering through biometric data to comply with other EU and member nation laws - specifically noting that police forces can sort images by hair or eye color to identify suspects.

Paragraph 31 addresses the prohibition of social scoring systems, which use AI to evaluate the trustworthiness of an individual:

AI systems providing social scoring of natural persons by public or private actors may lead to discriminatory outcomes and the exclusion of certain groups. They may violate the right to dignity and non-discrimination and the values of equality and justice. Such AI systems evaluate or classify natural persons or groups thereof on the basis of multiple data points related to their social behaviour in multiple contexts or known, inferred or predicted personal or personality characteristics over certain periods of time. The social score obtained from such AI systems may lead to the detrimental or unfavourable treatment of natural persons or whole groups thereof in social contexts, which are unrelated to the context in which the data was originally generated or collected or to a detrimental treatment that is disproportionate or unjustified to the gravity of their social behaviour. AI systems entailing such unacceptable scoring practices and leading to such detrimental or unfavourable outcomes should therefore be prohibited. That prohibition should not affect lawful evaluation practices of natural persons that are carried out for a specific purpose in accordance with Union and national law.

The People's Republic of China's Social Credit System helps to put individual debtors on blacklists, but is usually used to enforce regulations against companies.

The Act requires AI systems used for healthcare or employee recruitment to be monitored by humans, and ensure that they use high quality data. High risk AI systems will need to be registered in a database maintained by the EU and receive a declaration of conformity.

High risk AI systems will have to have a CE marking (physical or digital) to show that they conform with the Act. A CE 'conformité européenne' marking looks like this:

. . . it is used widely to show that the product conforms with health and safety regulations.

AI developers will not have to fully comply with the EU AI Act until August 2, 2027.

Relativity and AI - aiR for Review

Relativity's latest innovation is aiR for Review, which uses artificial intelligence to not only help find responsive documents, but also to describe why the document is relevant and cite to the parts of the document that make it relevant.

aiR is accessed in the Relativity mass operations menu.

. . . the litigation team enters a case summary using the same type of language you might use in a memorandum to describe the key players and key terms for a matter. A simple 2 to 5 sentence overview of the case should be provided:

The criteria to define exactly what should be regarded as relevant can be given as a description that you would send to another member of your team, rather than a series of keywords or sample documents.

When an individual document is viewed, the user can check a short summary of why it was found to be relevant, associated with the parts of the document that support this finding:

The process still requires a subject matter expert to not only devise the criteria used by aiR, but also to verify the initial results. A team of reviewers has to validate the findings generated by aiR on a statistical sample of the full document set.

When using aiR, Relativity recommends providing information clearly, but also as briefly as possible. The active voice should be used (so, "Mr. Johnson embezzled the bank's funds.", and not, "The bank's funds were embezzled by Mr. Johnson.") and double negatives should not be used. (Don't say, "Johnson can't hardly wait until the deal becomes final."). It is also necessary to avoid the qualifier, "including but not limited to".