Orion Labs Tech v. TalkDesk — Court Invalidates Six AI Bot Patents Under Alice but Spares Real-Time Translation Patent

Background

Orion Labs Tech sued TalkDesk for infringing seven patents related to AI-powered communication tools. The patents fell into four groups: the Intelligent Agent Patents (U.S. Patent Nos. 10,110,430; 10,462,003; and 10,924,339), which covered using intelligent agents or “bots” to perform services like recording, auditing, and transcribing communications within a group; the Bot Messaging Patents (U.S. Patent Nos. 10,897,433 and 11,127,636), which covered processing audio messages through voice libraries and natural language processing to produce “enhanced text” for bot execution; the ‘733 Patent (U.S. Patent No. 11,258,733), which covered using bots for transcription and delivery in group communications; and the Translation Patent (U.S. Patent No. 11,328,130), which covered real-time multilingual translation for group voice communications using a remote management server.

TalkDesk makes customer-service AI products including TalkDesk CX Cloud, Autopilot, Copilot, and Voice IVR. It moved to dismiss all seven patents as claiming ineligible subject matter under 35 U.S.C. § 101 — the Alice/Mayo framework that screens out patents claiming abstract ideas without an inventive concept.

The Court’s Holding

The court granted dismissal of six of the seven patents but denied dismissal of the Translation Patent, producing a split ruling that illustrates where the Alice line falls for AI-related patent claims.

Intelligent Agent Patents — ineligible. The court found these patents directed to the abstract idea of using a virtual assistant to perform services in a communication group. The claims described a generic “intelligent agent” — acknowledged to be synonymous with a “bot” — instantiated to perform recording, auditing, transcribing, annotating, and searching functions. The court noted that these are tasks humans have long performed (“a court reporter who transcribes the audio of a telephone conference call”) and that the patents’ purported advantages — that the bot is “immediately available,” “always functioning,” and “never fatigued” — are “inherent in the use of computers generally.” No inventive concept saved the claims, which recited only generic software modules and conventional computer components.

Bot Messaging Patents — ineligible. The court found these claims directed to the abstract idea of encoding and decoding information, specifically converting recorded audio to “enhanced text” for bot processing. The Federal Circuit has “consistently held that claims directed to encoding and decoding information or converting data from one format to another are abstract,” the court noted, citing directPacket Research v. Polycom and several other precedents. Orion Labs’ argument that the patents claimed “novel, purpose-built software functionality” failed because the claims recited purely functional limitations without specifying any concrete software configuration.

‘733 Patent — ineligible. The court applied the same reasoning: using a bot to deliver transcribed audio to an external platform is abstract, and the asserted innovation — maintaining source identity — was not actually claimed in the representative claim or asserted dependent claims.

Translation Patent (‘130) — survived. In contrast, the court found that the Translation Patent’s claims were not directed to the abstract idea of translation. These claims specified a concrete method: a remote management server that registers communication devices with language preferences and group settings, determines which devices should receive a speech input, identifies each device’s preferred language, groups devices by language, translates the speech input into each group’s preferred language, and sends the translated input. The court held that the claims “specify how the claimed result (i.e., ‘providing real-time translation for group communications’) is achieved” and rejected TalkDesk’s overgeneralized characterization. Because the claims survived Alice step one, the court did not reach step two.

Key Takeaways

• “Faster, better, always-on” is not enough for AI patents. Claiming that an AI bot automates tasks a human could perform — and doing so faster, without fatigue or distraction — recites advantages “inherent in the use of computers generally” and does not establish patent eligibility.
• Functional AI claims without implementation details are vulnerable. Patents that describe what a bot does (record, transcribe, translate) without specifying how the software achieves those results face strong § 101 challenges. The court repeatedly noted that Orion Labs’ claims were “functionally claimed without the necessary specificity or concreteness.”
• The Translation Patent survived because it specified a process. Claims that detail a specific server-side workflow — registering devices, logging language preferences, determining distribution groups, translating by language group — go beyond claiming the abstract idea of translation and describe a particular technical implementation. This distinction between “what” and “how” remains the key battleground for AI patent eligibility.
• Unasserted dependent claims may provide a lifeline. The court granted Orion Labs leave to amend its complaint to assert unasserted dependent claims of the Bot Messaging and ‘733 patents, suggesting that narrower dependent claims with more specific limitations might survive § 101 scrutiny.

Why It Matters

As AI-powered tools become ubiquitous in enterprise software — from chatbots to voice assistants to automated transcription — this ruling offers a clear roadmap for what makes an AI patent eligible versus abstract. Simply claiming a bot that performs human tasks in a communication environment is not enough; patent claims must specify the technical architecture and process by which the AI achieves its result. For companies building AI products, this decision underscores the importance of drafting patents that go beyond functional descriptions to claim specific computational workflows. The survival of the Translation Patent shows that AI-related claims can pass the Alice test — but only when they describe a concrete technical implementation rather than a general concept of AI automation.