AIs could, of course – and probably should – post disclaimers that highlight the risk that their output contains errors. Bard, for example, includes below the dialog, “Bard may display inaccurate or offensive information that does not represent the views of Google.” But such disclaimers do not immunize AI companies from potential defamation liability.
To begin with, such disclaimers cannot function as contractual waivers of liability: even if IAs users are deemed to waive their rights to sue based on misinformation when they expressly or impliedly acknowledge the disclaimers, which cannot waive the rights of the third who could be defamed.
The disclaimers also do not prevent the statements from being deemed actionable misrepresentations of fact. Defamation law has long treated false claims that could harm people’s reputations as actionable, even when there is a clear possibility that the claims are false. No newspaper can immunize itself from libel suits for a statement that “Our research reveals that John Smith is a paedophile” by simply adding “but be aware that this may be inaccurate” (much less by putting a line on the first page, “Warning: We may sometimes publish inaccurate information”). Likewise, if I write “I may remember incorrectly, but I remember that Mary Johnson was convicted of embezzlement”, it could be defamatory despite my disclaimer “I may remember- to be bad”.
This is reflected in many well-established doctrines on defamation. For example, “where a person repeats a slanderous accusation, even if the identification of the source or the indication that it is only a rumor, this constitutes a republication and has the same effect as the publication origin of calumny”.[1] When speakers identify something as a rumor, they implicitly say “it may be inaccurate” – but that doesn’t get them off the hook.
Indeed, according to the Restatement (Second) of Torts, “the republisher of a defamation or slander is liable to liability even if he expressly states that he does not believe the statement which he is repeating to be true” .[2] It is even clearer that a disclaimer that the statement only can being inaccurate cannot prevent liability.
Likewise, say that you are presenting both an accusation and the answer to the accusation. By doing this, you are making it clear that the prosecution “may [be] inaccurate.” This does not prevent you from being responsible for the repetition of the accusation.
To be sure, there are narrow and specific privileges that defamation law has developed to allow people to repeat possibly erroneous content without risk of liability, in particular contexts where such repetition is considered particularly necessary. For example, some courts recognize the privilege of “neutral reporting,” which immunizes impartial reporting of allegations and responses in certain situations: “[W]When a responsible and prominent organization … makes serious accusations against a public figure, the First Amendment protects accurate and disinterested reporting of those accusations,” even when the reporter has serious doubts about the accuracy of the accusations.[3] But other courts reject privilege altogether.[4] And even those who accept it only apply it to narrow situations: reporting allegations and responses remains actionable – even if the report makes it clear that the allegations may be wrong – when the allegations relate to issues of private interest or are made by persons or entities that are not “responsible” and “prominent”.[5] He certainly remains liable to prosecution when the allegations themselves are repeated or reported in error by the speaker.
Privilege is considered necessary precisely because of the general rule that, in the absence of such privilege, the transmission of claims may be defamatory even when it is clearly stated that the claims may be in error. And the privilege is a narrow exception justified by the “fundamental principle” that, “when a responsible and prominent organization … brings serious charges against a public figure”, the media should be able to engage in “accurate and disinterested reporting on these charges, “because they are ‘newsworthy’ simply because ‘they were made’.[6]
Similarly, the restricted rumor privilege allows a person to repeat certain types of rumors to particular people to whom they have a particular duty – such as friends and family members – if the rumors relate to likely conduct. to threaten these people. (This stems from what is seen as the special legitimacy of people protecting the interests of their friends.[7]) That’s why, for example, if Alan tells Betty that he heard a rumor that Betty’s employee, Charlie, was a thief, Alan is absolved of any responsibility.[8] But the privilege exists precisely because, without it, conveying factual allegations to (say) an outsider or to the general public – even recognizing that they “may [be] inaccurate” – actionable.[9]
Now a disclaimer that actually describes something like fiction, or as parody or as hypothetical (both forms of fiction), can very well be effective. Recall that, in defamation cases, a “key question is whether the impugned expression, however labeled by the defendant, would reasonably appear to state or imply assertions of objective facts”.[10] A statement that obviously contains no factual assertion – as opposed to merely mentioning a factual assertion about which the speaker expresses uncertainty or even disbelief – is not actionable.[11] But neither ChatGPT nor Bard actually describe themselves as producing fiction, because that would be a bad business model for them. Rather, they All their general reliability, and simply recognize the risk of error. This recognition, as the cases discussed above show, does not exclude liability.
[1] Ringler Associates Inc. v. Maryland Casualty Co., 80 Cal. App. 4th 1165, 1180 (2000).
[2] Restatement (Second) of Torts § 578 cmt. e; see also Martin v. Wilson Pub. Co., 497 A.2d 322, 327 (RI 1985).
[3] Edwards v. National Audubon Soc’y, 556 F.2d 113 (2d Cir. 1977). A few later cases extended this to certain charges on matters of public interest against private figures. Others have rejected the privilege for statements about private figures, without commenting on its availability to public figures. See, for example, Khawar c. Globe Int’l, Inc., 965 P.2d 696, 707 (Cal. 1998); Fogus v. Cap. Cities Media, Inc., 444 NE2d 1100, 1102 (App. Ct. Ill. 1982).
[4] Norton v. Glenn, 860 A.2d 48 (Pennsylvania 2004); Dickey v. CBS, Inc., 583 F.2d 1221, 1225–26 (3d Cir.1978); McCall v. Courier-J. & Louisville Times, 623 SW2d 882 (Ky. 1981); Postill c. Booth Newspapers, Inc., 325 NW2d 511 (Mich. App. 1982); Hogan v. Herald Co., 84 AD2d 470, 446 (NY App. Div. 1982).
[5] A few authorities have applied this privilege to the specific disclosure of allegations on matters of general public interest, but this appears to be a minority rule. Barry v. Time, Inc., 584 F. Supp. 1110 (ND Cal. 1984); Texas Civil. Code § 73.005.
[6] Edwards, 556 F.2d at 120. Similarly, the privilege of fair reporting allows allegations that have been made in governmental proceedings to be repeated accurately, because of the deeply rooted principle that the public should be able to know what has been said in these proceedings, even when such statements damage reputation. But it too is strongly limited to accurate repetition of allegations originally made in government procedure.
[7] Restatement (Second) of Torts § 602.
[8] Identifier. cmt. 2. Another classic illustration is of a parent warning an adult child of a rumor that the child’s future spouse or lover is untrustworthy. Identifier. cmt. 1.
[9] See, for example, Martin c. Wilson Pub. Co., 497 A.2d 322, 327 (RI 1985).
[10] Takieh v. O’Meara, 497 P.3d 1000, 1006 (Ariz. Ct. App. 2021).
[11] See, for example, Greene c. Paramount Pictures Corp., 813 F. App’x 728, 731–32 (2d Cir. 2020). Even so, a court might allow liability if it finds that a reasonable person who knows the plaintiff would understand that the defendant’s alleged fiction is in fact meant to be as novel key which transmits factual statements about the applicant. The presence of a warning would then not be decisive. See, for examplePeter c. Griffin, no. 20-CV-1173-PB, 2021 WL 4477764, *6 n.10 (DNH 30 Sept. 2021).