UKIPO finds AI content-moderation system not shown to infringe patent

The UK Intellectual Property Office (UKIPO) has concluded that an AI-powered content-moderation system offered by Unitary Ltd. has not been shown to infringe a patent owned by Lighthouse Technologies Limited.

In Opinion 11/26, issued on September 23, the UKIPO considered whether Unitary’s content-moderation system infringed independent claims 1 and 8 of GB2602159 B, Method and system for redacting undesirable digital content. The patent was filed on December 21, 2020, granted on January 18, 2023, and remains in force.

The opinion was issued under section 74A of the Patents Act 1977 following a request by Lighthouse. Cooley (UK) LLP filed observations on behalf of Unitary. The request concerned only claims 1 and 8; no infringement was alleged in relation to the dependent claims.

AI content moderation at the center of dispute

The patent concerns a computer-implemented method and system for redacting undesirable digital content, including image and video content.

Claim 1 requires, among other things, an identification server to receive a content request originating from a user device, identify an associated user account, and tag the request with a restriction identifier indicating a restriction level associated with the user.

An inspection engine then analyzes the requested visual content before it is transmitted to the user device. The claim requires the inspection engine to use an AI model trained through machine learning to determine whether the content is appropriate for the user based on the restriction level.

The analysis result must include both a score associated with the visual content and a confidence level output by the inspection engine, with both taken into account when determining whether the content should be redacted. If it is classified as undesirable, the claimed process requires the content to be redacted and a redacted or censored version transmitted to the user device.

Claim 8 is the corresponding system claim, and the UKIPO said that the substantive infringement issues were essentially the same for both claims.

UKIPO focuses on architecture and user flow

A central issue was how the claimed system’s user-related workflow compared with the Unitary system.

The evidence before the UKIPO described Unitary as providing B2B content-moderation solutions. Customers submit user-generated content to the system for classification according to policies specified by the customer.

In one version, Unitary returns a score indicating the classification, and the customer decides what action to take. In another, Unitary performs a pre-approved action determined by the customer, but that action takes place on the customer’s system rather than Unitary’s system.

The UKIPO found that the patent required a particular logical flow of content from a provider to an individual user.

Although the patent does not prescribe a particular physical location for every component of the system, the UKIPO concluded that the claims require a logical flow in which a user requests content and that particular user’s restriction level is relevant to the subsequent processing.

It therefore rejected Lighthouse’s argument that the claims could encompass content being pre-processed for an unknown future user and subsequently supplied to that user.

The UKIPO said the claimed inspection engine must be aware of the end user, meaning that pre-processing content before a particular user requests it did not fall within the claims as construed.

Restriction level was present, but at a different stage

The UKIPO did not, however, reject every aspect of Lighthouse’s infringement case.

It concluded that Unitary’s customer-specific and use-case-specific configuration could amount to a restriction level for the purposes of the claims. But that restriction level arose at a secondary stage, when a content provider delivers content to a user, rather than through the claimed sequence in which the user’s content request is received and processed.

The distinction was therefore one of functional flow and timing, rather than simply whether Unitary’s system used policies or thresholds to determine what content could be supplied.

Score and confidence requirements not established

The UKIPO separately considered the patent’s requirement for both a score and a confidence level.

Lighthouse relied on Unitary’s API documentation and an AWS case study. The UKIPO accepted that the documentation showed that Unitary’s system generated a score and that the score might have some relationship to confidence.

However, it was not persuaded that the evidence established that a separate confidence level was calculated and used alongside the score in determining whether visual content should be redacted, as required by the claims.

The Office noted that the safety score could potentially be a composite of prevalence and confidence, or could be correlated with confidence without a separate confidence calculation being performed. The available documentation did not make the relevant relationships sufficiently clear.

The UKIPO also considered the patent’s redaction requirement.

While Unitary’s documentation described thresholds that could be used to block content, the evidence did not establish that use of the system would result in a redacted or censored image or video being supplied to the user. Blocking could instead result in no response or another action on the user’s computer, and the evidence did not demonstrate the specific redaction mechanism required by the claims.

Contributory infringement also not established

Lighthouse alternatively relied on contributory infringement under section 60(2) of the Patents Act 1977.

The UKIPO considered whether the required steps of the patented invention would follow directly and automatically from use of Unitary’s system.

It concluded that the evidence did not establish this. The adaptations necessary to implement the claimed invention were not necessary to implement the Unitary system, and it was not clear that a customer using Unitary’s system would implement it in the manner required by claim 1.

The UKIPO therefore was not persuaded that the evidence established contributory infringement.

Actavis doctrine of equivalents rejected

Lighthouse also argued that Unitary’s system infringed under the doctrine of equivalents, relying on the three questions established by the UK Supreme Court in Actavis UK Ltd v. Eli Lilly and Company [2017] UKSC 48.

The UKIPO accepted that Unitary’s system could produce a similar overall result — suppression of undesirable visual content — but concluded that the systems’ architectures were significantly different.

The patent was directed to a process for controlling delivery of content requested by an individual user, whereas the Unitary system involved a different architecture in which content was submitted for classification, and the resulting decision could be acted upon by the customer.

The UKIPO concluded that the Unitary system therefore did not achieve the result in substantially the same way as the patented invention. It likewise concluded that a skilled person would regard the architectures as substantially different.

The Office consequently found no infringement under the doctrine of equivalents.

UKIPO opinion stops short of a court finding

The UKIPO’s conclusion was carefully framed.

It stated: “the Unitary content moderation system has not been shown to infringe claims 1 and 8 of the Patent.”

The Office said this conclusion was reached after considering direct infringement under section 60(1), contributory infringement under section 60(2), and equivalence under the Actavis questions. The opinion does not amount to a finding following fully litigated court proceedings that Unitary’s system cannot infringe the patent.

The UKIPO expressly states that the opinion is based on the material that the parties chose to put before it.

Lighthouse can also apply for a review of the opinion under section 74B of the Patents Act 1977 and rule 98, with the application required within three months of the opinion’s issue.

AI patent infringement tested feature by feature

The opinion provides a useful example of how infringement of an AI-related patent can turn on the specific architecture and data flow claimed, rather than simply on the use of similar AI functionality.

Both the patent and Unitary’s product concern AI-assisted content moderation. But the UKIPO’s analysis focused on the precise relationship between the user, the content request, restriction level, inspection engine, scoring and confidence information, and the subsequent treatment of the content.

For AI patent owners, the opinion illustrates the importance of being able to establish how an accused system implements each relevant claim feature. For potential defendants, it demonstrates how differences in system architecture and the way information moves through an AI-enabled platform can become central to an infringement analysis.

The UKIPO’s official records confirm that Opinion 11/26 was issued on September 23 and concerned infringement of GB2602159.

← Back to News