US 12299603

I'll search for this specific patent number and related litigation docket information.

The initial searches returned noise. Let me try more targeted queries.

I found relevant litigation. Let me dig deeper into the OpenAI lawsuit, the specific patents, and any CAFC docket.

US Patent 12,299,603 — Analyst Summary

Important date caveat up front. You gave today's date as April 26, 2026, but my tool environment and the retrieved sources reflect late September 2026. The material I found includes events that post‑date April 26, 2026 (see litigation below). I flag this because it materially affects what "current" means here. Per my operating rules, I am treating the live search results as ground truth and noting the conflict rather than silently reconciling it.

1. Bibliographic data (from the authoritative full text provided)

Note on dates: This is a continuation-style filing with a 2011 priority claim but a 2024 filing date; the term runs from the 2011-10-07 non-provisional, hence the ~2031 expiration. So the "filing date" that matters for prior art is the 2011 family, not the 2024 paper filing.

2. Litigation / docket search results

What I found (district court):

• Many Worlds 2T Innovations LLC v. OpenAI OpCo, LLC, U.S. District Court for the Eastern District of Texas, Marshall Division, Case No. 2:26-cv-00774, complaint filed August 31, 2026 (assigned to Judge Rodney Gilstrap). Sources: ipwatchdog.com (Sept 2, 2026); ailawsuittracker.com; PACER Monitor case 66548709.
• The '603 patent is one of five patents-in-suit: U.S. Pat. Nos. 8,676,742; 8,843,433; 10,699,202; 12,307,388; and 12,299,603.
• Accused products/services (as reported): ChatGPT, ChatGPT Memory, ChatGPT Shopping Research, ChatGPT Deep Research, ChatGPT Projects, ChatGPT Search, and OpenAI Vector Stores / File Search.
• The complaint reportedly asserts direct and induced infringement, seeks a jury trial, past and continuing royalties, enhanced (willful) damages, and an injunction. It also reportedly anticipates a §101 eligibility challenge by OpenAI.
• For the '603 patent specifically, the IPWatchdog account describes the theory as targeting OpenAI's retrieval stack (embeddings/vector retrieval), but that sentence is truncated in the source I retrieved, so I cannot quote the full allegation.

Regarding CAFC 2026 dockets: I searched for a Federal Circuit docket tied to patent 12,299,603 and found none. As of the sources I retrieved, this patent appears only in the pending EDTX district-court action — no appeal, no IPR appeal, and no CAFC briefing identified. I cannot rule out a docket I did not surface, so treat this as "no CAFC activity found," not "confirmed none."

3. Subject matter / abstract

Caution: The authoritative text supplied to me is the specification/description body (Google Patents rendering, starting at the "This invention relates to…" definitions block). It is truncated (it cuts off mid-sentence in the "self-profiling" discussion of Table 1) and does not include a verbatim abstract or the claims section. The following is therefore grounded in the specification and the IPWatchdog description, not a quoted abstract:

The patent is directed to a vector-based search method and system that combines behavioral indexing with content (search-engine) indexing inside an "adaptive" / learning layer built on a fuzzy content network (nodes related by degrees of affinity rather than binary links). Content objects are indexed by a search engine into object contents vectors (OCVs) — constituent elements (words/phrases/concepts) with relevance weights. Behavioral information generates affinity vectors such as MTAV (member-to-topic), MTEV (member-topic expertise), CTAV/OTAV (content-to-topic), CCAV (content-to-content), TTAV (topic-to-topic), and MMAV (member-to-member). A search or navigation context is used to establish a contextual neighborhood, and candidate objects are selected by comparing the user's MTAV/MTEV against CTAVs (behaviorally indexed sources) and/or against OCVs (contents-indexed sources), with optional quality filtering and a learning-layer API for programmatic calls. This is stated to address "cold-start" and black-box explanation problems relative to collaborative filtering.

4. Independent claims — overview (with explicit uncertainty)

I do not have the authoritative claim text. The full text provided to me stops before the claims, and my searches did not retrieve a verbatim claim set for US 12,299,603. I will not fabricate claim language. What follows is a probable functional overview inferred from the title, the specification's emphasis, and the litigation reporting — and should be verified against the issued claims on USPTO Patent Center / Google Patents before any reliance.

Probable independent claim coverage (subject to verification):

1. (Method) A computer-implemented vector-based search method: content-indexing objects to produce weighted object contents vectors; applying a search request against those vectors to identify objects; and supplementing/augmenting the search results using behaviorally-derived affinity vectors (e.g., CTAV) to surface additional objects the pure content search would miss — i.e., hybrid content-search + behavioral-affinity retrieval.
2. (System) A system with processor/memory implementing the same: a content-indexing engine producing OCVs, a behavioral-indexing engine producing affinity vectors, and a retrieval function that returns a contextual neighborhood based on both, optionally with user-personalization via MTAV/MTEV and quality filtering.
3. (CRM / non-transitory medium) A computer-readable medium storing instructions to perform the vector-based search and affinity-supplemented retrieval steps.

Confidence: low-to-moderate on the exact claim scope; moderate that the claims center on combining OCV-based retrieval with behaviorally-derived affinity-vector retrieval and a defined contextual neighborhood.

5. Summary of uncertainty

• Verbatim abstract and claim set of US 12,299,603 were not available in the provided text or my searches — flagged, not invented.
• Date conflict (your stated April 26, 2026 vs. my environment/sources at ~Sept 27, 2026) materially affects the litigation timeline — the OpenAI suit and the assignee reassignment are relevant to the patent's current posture.
• No CAFC 2026 docket was found for this patent; only the EDTX action (2:26-cv-00774) was identified.
• Assignee is now Many Worlds 2T Innovations LLC (reassignment from ManyWorlds, Inc.), which is the plaintiff in the asserted litigation.

Suggested verification steps (I could not complete them within the search budget): pull the issued claims directly from USPTO Patent Center for application 18/754,550, and confirm the '603 count in the EDTX complaint via PACER/RECAP docket 2:26-cv-00774.

I'll research litigation involving US patent 12299603 across multiple sources.

Initial searches didn't surface direct hits. Let me dig deeper into the owner and litigation databases.

Two cases appear to involve this patent. Let me verify which patents are asserted in each and check for additional cases.

Litigation Involving U.S. Patent No. 12,299,603

⚠️ Preliminary note on the date discrepancy

Your prompt states the current date is April 26, 2026. However, the primary records I retrieved (including the authoritative patent text at patents.google.com and the litigation reporting below) are dated September 2026, and the OpenAI action discussed below was filed August 31, 2026 — after the date stated in your prompt. I am reporting the search results as the current ground truth per your strict rule, but flag that one case postdates the date given in your instructions. Please verify against live PACER/Unified Patents records before relying on this.

Patent identification

• U.S. Patent No. 12,299,603 B2 — "Vector-Based Search Method and System"
• Application No. 18/754,550; filed 2024-06-26; granted/issued 2025-05-13; priority date 2011-03-29; anticipated expiration 2031-10-07.
• Inventors: Steven Dennis Flinn; Naomi Felina Moneypenny.
• Original assignee ManyWorlds, Inc.; assignment to Many Worlds 2T Innovations LLC recorded (reel/assignment dated 2026-04-15 per the Google Patents record).
• The Google Patents family record carries a "Family has litigation" / "First worldwide family litigation filed" flag (Darts-IP family 47142579), consistent with the cases below.

Known litigation

Case 1 — Many Worlds 2T Innovations LLC v. OpenAI OpCo, LLC

The '603 patent is the fifth of five asserted patents and is described as covering the "vector-based search method and system" — the complaint allegedly maps it to OpenAI's retrieval stack (vector-store retrieval indexes organizing embedded content by degrees of relatedness; relevance thresholds, max-result settings, reranking, and hybrid semantic-keyword search).

Source grounding: IPWatchdog, "OpenAI Sued Over AI Patents Covering Retrieval, Personalization and Content Generation" (Sept. 2, 2026), https://ipwatchdog.com/2026/09/02/openai-sued-over-ai-patents-covering-retrieval-personalization-content-generation/; PacerMonitor docket https://cdn.pacermonitor.com/public/case/66548709/Many_Worlds_2T_Innovations_LLC_v_OpenAI_OpCo,_LLC; AI Lawsuit Tracker, https://ailawsuittracker.com/cases/many-worlds-2t-innovations-llc-v-openai-opco-llc-2-26-cv-00774/.

Case 2 — Many Worlds 2T Innovations LLC v. Spotify Technology S.A., Spotify AB, and Spotify USA Inc.

I include this case because (a) it is the same plaintiff and patent family, (b) the plaintiff name "2T" appears tied to the accused "Two Tower" recommendation engine, and (c) it may later be amended to assert the '603 patent — but as of the records retrieved, the '603 patent is not confirmed as asserted here.

Source grounding: PacerMonitor docket, https://cdn.pacermonitor.com/public/case/64166196/Many_Worlds_2T_Innovations_LLC_v_Spotify_Technology_SA_et_al; Dockets.Justia E.D. Tex. patent filings index; Ex Parte AI Lab complaint analysis, https://ai-lab.exparte.com/case/dct/txed/2:26-cv-00303/doc/analysis/1.

Negative findings

• No PTAB proceedings (IPR/PGR/CBM) specifically challenging U.S. 12,299,603 were identified. Note the patent issued 2025-05-13, so the one-year IPR window for an accused infringer (35 U.S.C. § 315(b)) is still running relative to the OpenAI and Spotify suits.
• No Federal Circuit or other appellate activity involving the '603 patent was found.
• I did not find any ITC Section 337 investigation naming the '603 patent.
• I could not directly query the Unified Patents litigation portal (https://portal.unifiedpatents.com/litigation/caselist) or PACER within the constraints of this session; the results above come from PacerMonitor, Dockets.Justia, IPWatchdog, and an AI litigation tracker.

Bottom line

As of the records retrieved, U.S. Patent No. 12,299,603 has been asserted in one confirmed patent infringement action — Many Worlds 2T Innovations LLC v. OpenAI OpCo, LLC, No. 2:26-cv-00774 (E.D. Tex., filed Aug. 31, 2026), which remains pending with no outcome. The related Spotify action (2:26-cv-00303, E.D. Tex., filed April 17, 2026) is a same-plaintiff suit but I could not confirm the '603 patent is among its asserted patents.

Caveat: Litigation dockets change quickly, and this appears to be an active assertion campaign by Many Worlds 2T Innovations LLC. Please re-verify against live PACER/Unified Patents records (and the Darts-IP family link on the Google Patents page) before relying on this list — particularly given the date conflict noted at the top.

I'll verify the structured ODP data against live sources before writing anything.

PTAB Proceedings — US 12,299,603 ("Vector-Based Search Method and System")

Proceedings overview

Total AIA trial proceedings on file: 0 — 0 active, 0 claims invalidated, 0 claims sustained, 0 settled, 0 institution denials. The structured "PTAB proceedings on file" block (USPTO ODP) returns an empty list, and my independent web searches surfaced no IPR, PGR, or CBM naming U.S. Patent No. 12,299,603. The bottom line for a defendant: there is no PTAB record to exploit and no estoppel to rely on — the '603's claims are 100% untested in an AIA forum, so any invalidity theory you want to run, you have to build and file yourself.

This is a "blank slate" posture, which is materially different from both "hardened by surviving IPRs" and "claims canceled." Do not let anyone tell you the patent has been "PTAB-validated" — institution or an FWD for the '603 simply does not exist.

Per-proceeding detail

No AIA trial proceeding exists to chart. The template below is unfilled because there is no proceeding number, petitioner, or paper to cite — and I will not invent one.

(none) — no petitioner v. Many Worlds 2T Innovations LLC

• Type: n/a (no IPR, no PGR, no CBM)
• Filed: n/a
• Status: n/a — the ODP canonical list is empty; no web-surfaced counter-evidence
• Judge panel: n/a
• Petition grounds: n/a
• Institution decision: n/a
• Final Written Decision: n/a
• Settlement / termination: n/a
• Appeal: n/a
• Defensive value: none derivable — there is no PTAB paper to quote, no claim held unpatentable to lean on, and no § 315(e)(2) estoppel binding any party.

Search basis (so you can audit the negative). I ran targeted queries for "12,299,603" + IPR/PTAB/OpenAI, "12299603" IPR, ManyWorlds PTAB, and "Many Worlds 2T Innovations" IPR. The "12,299,603" + IPR query returned zero results; the remainder returned only lookalike or unrelated matters (below). I could not re-run a third round because the tool-call budget for this session was exhausted, so treat the negative as strong but not exhaustive.

Lookalike / adjacent matters examined and ruled out — none of these involve the '603:

Litigation context (not PTAB, but it drives the PTAB timeline)

The '603 is freshly asserted, which explains the empty PTAB docket:

• 2025-05-13 — '603 granted (application 18/754,550, filed 2024-06-26; continuation in the Flinn/Moneypenny 2011 family; stated priority 2011-03-29).
• 2026-04-15 — Assignment recorded from ManyWorlds, Inc. to Many Worlds 2T Innovations LLC — a litigation-branded vehicle ("2T" = the two-tower recommender architecture targeted in the Spotify case).
• 2026-04-16 — Many Worlds 2T Innovations LLC v. Spotify Technology S.A. et al., E.D. Tex. 2:26-cv-00303 (patents listed include '742, '829, '433, '729, '202; I could not confirm whether the '603 is among the additional patents asserted). Docket Control Order sets Markman for 2027-11-16 and jury selection for 2028-05-08.
• 2026-08-31 — Many Worlds 2T Innovations LLC v. OpenAI OpCo, LLC, E.D. Tex. (Marshall) 2:26-cv-00774, asserting U.S. Pat. Nos. 8,676,742; 8,843,433; 10,699,202; 12,307,388; and 12,299,603. Press coverage notes the complaint pre-emptively addresses § 101, i.e., the patentee expects an Alice challenge. (IPWatchdog)

Statutory windows for a '603 challenger (verify dates against actual service, not filing):

• § 315(b) one-year bar: for OpenAI, service of the 2026-08-31 complaint starts a ~12-month clock — realistically the first '603 IPRs would land in the mid-2027 window and would not produce an FWD until roughly 2028–2029.
• PGR window (§ 321(c)): 9 months from grant = on or about 2026-02-13. Closed. Note a threshold wrinkle: if all '603 claims carry pre-2013-03-16 effective filing dates (the family traces to 2011), PGR was never available at all. I could not verify the claim-by-claim priority chain from the materials provided.
• CBM: unavailable — the transitional program sunset 2020-09-16; the '603 issued in 2025.

Strategic summary

Claim status. Nothing is CANCELED, nothing is SUSTAINED, everything is UNTESTED. Because no AIA trial exists, there is no claim-level adjudication to quote and — importantly — I cannot even enumerate the '603 claim set from the patent text retrieved for this analysis; the fetched document included the specification's Definitions/summary material but not the claims. Any statement about "claims 1–N" of the '603 would be fabrication, so I make none. The only claim-level assertion data point I found (claims "1, 15, and 17" in a complaint-analysis snippet tied to the Spotify case) is not reliably attributable to the '603 and should be verified against the actual pleadings on PACER before it is used for anything.

Estoppel landscape. No petitioner has any § 315(e)(1) or § 315(e)(2) estoppel, because no petitioner has ever been in an AIA trial on this patent. That means the entire prior-art universe remains open to a defendant — there are no grounds "raised or reasonably could have been raised" by anyone. The corollary risk is § 325(d): the '603 was examined in 2024–2025 over a Flinn/Moneypenny specification that has been before the Office for two decades, so a petitioner should expect discretionary-denial arguments and should build the record around art that was not before the examiner (e.g., non-patent literature, product/system prior art, and art directed at the two-tower/embedding-retrieval architecture the patentee now accuses).

Pattern signals. No repeat petitioner, no follow-on filings, and no defensive aggregator (Unified Patents, RPX, etc.) found anywhere in the '603 chain — the IPR2021-01557 hit was a party-name coincidence, not a connection. The patent owner has not pursued any PTAB appeal for this patent (there is nothing to appeal). What does exist is a broad, multi-defendant assertion campaign across the same 2011 family (OpenAI, Spotify), which is the classic precondition for IPRs: well-asserted, recently-issued continuations reliably attract them once the § 315(b) clock matures.

One adjacent signal worth flagging (not a PTAB proceeding). Two ex parte Board appeals involving Flinn-named applications in the same general portfolio went badly on § 101:

• Appeal 2018-005988 (Application 13/269,958) — the Board affirmed that claim 7 was directed to an abstract idea ("certain methods of organizing human activity"), rejecting the argument that a generically recited mobile device and "computer-implemented neural network" supplied an inventive concept. (decision)
• Appeal 2019-001219 (Application 14/497,645) — the Board adopted the Examiner's § 101 rejection of claims 1–3, 5–11, and 14–20, while reversing § 102/§ 103 rejections. (e-FOIA)

These are ex parte appeals, not AIA trials, and I could not confirm that either application sits in the '603's direct priority chain — treat them as posture/drafting signals about this patent family's § 101 exposure, not as invalidity rulings against the '603.

Recommended next steps

1. Confirm the negative directly in the Office before you rely on it. Search PTAB E2E / PTAB Center for Patent No. 12299603 and check the "Proceedings" and "Related Matters" tabs (https://ptab.uspto.gov; decisions at https://www.uspto.gov/patents/ptab/decisions). Also check the Darts-IP "family litigation" link on the Google Patents family page — a litigation flag is not an AIA-trial flag, and the two are commonly conflated. Because the ODP ingest can lag, an E2E search is the only way to convert my "strong negative" into a "confirmed negative."
2. If you are OpenAI (or a co-defendant served on the 2026-08-31 complaint): your IPR clock is running. Calendar the § 315(b) deadline from the service date and start prior-art mining now. There is no estoppel to worry about and no FWD to cite — the FWD you'll be linking to in two years does not exist yet, so the value here is purely offensive.
3. Weight § 101 heavily, in the right forum. Given the family's § 101 appellate history and press reporting that the complaint itself anticipates an Alice challenge, a § 101 motion in E.D. Tex. may deliver invalidation faster and cheaper than an IPR (PTAB cannot decide § 101 at all in an IPR — § 311(b) limits IPRs to § 102/§ 103 on patents and printed publications).
4. Note the term/economics. Google Patents lists an anticipated expiration of 2031-10-07 (measured from the 2011-10-07 non-provisional, before any PTA/PTE). If that holds, you are litigating a patent with roughly five years of nominal life — which materially affects settlement math and may favor a district-court § 101/§ 112 strategy over a multi-year IPR campaign.
5. Do not cite any claim number of the '603 without pulling the issued claims. Nothing in this record establishes which claims exist or which are asserted; the complaint-analysis snippet about claims 1, 15, and 17 needs PACER verification.

Bottom line: no PTAB activity exists on US 12,299,603. The absence is itself the signal — a freshly issued 2025 continuation now being asserted against OpenAI and Spotify is exactly the profile that attracts its first IPR within 12–18 months of service. The paper you'd want to quote doesn't exist yet; the window to create it is open.

I'll research this patent's ownership chain across the specified sources.

Let me dig into the assignment records, the original assignee's history, and the asserting entity.

Ownership Reconstruction — US 12,299,603 B2 ("Vector-based search method and system")

Verification starting points

• Google Patents legal events (authoritative for the events Google surfaces): https://patents.google.com/patent/[US12299603B2](/patent/US12299603B2)/en#legalEvents
• USPTO Assignment Center search: https://assignmentcenter.uspto.gov/ and the legacy mirror https://assignment.uspto.gov/patent/index.html
• USPTO Patent Center for application 18/754,550

Data-integrity note up front. I was able to confirm the events in the chain (assignor, assignee, conveyance, dates) but I could not retrieve the reel/frame numbers or the recording correspondent for the 2026 assignment from the sources reachable in this session. Rather than invent them, I have marked those fields NOT VERIFIED and stated exactly where they must be pulled (Assignment Center, which is the only place that exposes reel/frame + correspondent). Everything else below is grounded in the sources cited inline.

Inventors

Inventor addresses on the face of related family members (e.g., US 9,454,730 B2, assignee "ManyWorlds, Inc., Brenham, TX") confirm the employment relationship.

Unusual-pattern check — negative. There is no evidence of inventors departing the original assignee within 12 months of filing. The opposite is true: Flinn and Moneypenny are the continuous inventors of record on this family from the 2011-03-29 provisional through the 2024-06-26 continuation filing (a ~13-year inventor tenure). The "inventors bailed out pre-fire-sale" tell is not present.

Original assignee

ManyWorlds, Inc. (also rendered "ManyWorlds Inc" in the legal events) — Texas corporation, historically addressed at Brenham, TX and Houston, TX (510 Bering Dr, Houston).

• Line of business: Adaptive/personalization software and management consulting. Directory data describes ManyWorlds as a Houston software company + business management consultancy, founded 1998, ~25–100 employees, revenue band $1M–$10M, with Steven Flinn as CEO (allbiz.com business profile). This is the entity credited with early "adaptive systems" / fuzzy-network personalization work described in the specification.
• Did they ship a product embodying the claims? Partially / historically. ManyWorlds commercialized adaptive recommendation and expertise-discovery software, which is what the specification describes. It was a small software-and-consulting shop, not a platform-scale product company. I found no evidence of a currently marketed product embodying the claims.
• Current status: Unclear / no public dissolution or bankruptcy found. The entity is still alive enough to be the assignor of record on a 2026 transfer, so it was not dissolved before 2026. It does not appear to be an active operating competitor of the accused parties. I found no Chapter 7/11 filing for ManyWorlds, Inc. — so the "bankruptcy fire-sale" theory is unsupported.

Assignment timeline

The chain has two links, one of which is a post-issuance transfer. Google Patents surfaces exactly one reassignment record for this patent:

• 2026-04-15 (recorded) — Reel/Frame NOT VERIFIED

  • Conveyance: Assignment of Assignor's Interest ("ASSIGNMENT OF ASSIGNOR'S INTEREST" per the Google Patents legal-events entry; recorded as a reassignment)
  • Assignor: MANYWORLDS, INC.
  • Assignee: MANY WORLDS 2T INNOVATIONS LLC
  • Correspondent: NOT VERIFIED — not exposed by the sources reachable here. This is the single most important missing field; see the flag under Signal 3.
  • Context: Transfer-to-asserter. The whole right, title and interest moved from the original operating assignee to a litigation-oriented LLC; within two days of recordation the LLC filed its first infringement suit (Spotify, 2026-04-17).
  • Source: Google Patents legal events, https://patents.google.com/patent/US12299603B2/en (event dated 2026-04-15: "Assigned to MANY WORLDS 2T INNOVATIONS LLC … Assignors: MANYWORLDS, INC.").

• 2024-06-26 (application filing) — inventor → ManyWorlds, Inc. Reel/Frame NOT VERIFIED

  • Conveyance: Assignment (employment/obligation-to-assign), implicit in the application being filed by ManyWorlds Inc and published/granted with ManyWorlds as assignee.
  • Assignor: Flinn and Moneypenny | Assignee: ManyWorlds, Inc.
  • Context: Internal / employment assignment (original acquisition of title by the operating company). Treat the reel/frame as unverified; confirm in Assignment Center.

Ruled out by the record: no security agreement, no merger, no change-of-name, no release, no correction, no intermediate chained LLC. This is not a cascading multi-hop chain — it is one operating company and one downstream LLC.

Timeline diagram

timeline
    title Ownership of US 12299603
    2011 : Provisional application filed
         : Two parent applications filed
    2024 : Continuation filed by ManyWorlds Inc
    2025 : Patent granted May 13
    2026 : Reassigned to Many Worlds 2T Innovations
         : Spotify suit filed Apr 17
         : OpenAI suit filed Aug 31

NPE / troll-pattern signals

1. Shell-entity transfer — present (strong on the transfer; incomplete on the corporate tells)

The patent moved from operating assignee ManyWorlds, Inc. to Many Worlds 2T Innovations LLC per the assignment recorded 2026-04-15. The LLC's only observable activity in the record is filing patent infringement suits (two in 2026); I found no products or services offered by Many Worlds 2T Innovations LLC. The name encodes the accused architecture — "2T" = Spotify's "Two Tower (2T) recommendation engine" named in the complaint (Compl. ¶55, as summarized by Ex Parte's complaint analysis, https://ai-lab.exparte.com/case/dct/txed/2:26-cv-00303/doc/analysis/1). Not verified: the LLC's state of formation, registered-agent address, and single-membership — those require the Assignment Center cover sheet / state registry and should be pulled before this is called a textbook shell.

2. Known asserter in the chain — not present (as to the named lists); unclear as to aggregator directories

Neither assignee matches the enumerated public NPE lists (Acacia, Marathon, Intellectual Ventures, IPNav, Wi-LAN, Mosaid/Conversant, Vringo, Pendrell, Innovatio, MPHJ, Lumen View, Round Rock, Document Generation Corp, Spangenberg entities). Many Worlds 2T Innovations LLC appears to be a new, single-family asserter rather than a member of those established portfolios. I could not confirm whether it appears in the Unified Patents or RPX high-frequency-plaintiff directories — treat as unclear and check those directories directly.

3. Repeat correspondent across the chain — unclear / NOT VERIFIED

There is only one post-issuance link, so recurrence within this chain is impossible to establish. Critically, the recording correspondent for reel/frame on the 2026-04-15 assignment is not exposed by the sources reachable here, and I will not guess it.

• Do not conflate the assignment correspondent with the litigation counsel of record, who are visible on the dockets: Kevin Cadwell, Bradley Bowling, Lisa Maria Thomas, Mark Douglas Siegmund, Jack Shaw, William David Ellerman, Gregory Phillip Love, Ryan Scott Loveless, Shuya Yang (see PacerMonitor dockets for 2:26-cv-00303 and 2:26-cv-00774). These are prosecution/litigation appearances, not proof of who recorded the assignment.
• Action item: the highest-value follow-up on this patent is pulling the reel/frame + correspondent from Assignment Center and cross-checking that name against other ManyWorlds-family recordings. The shell changes; if a single recording attorney (or firm) filed the ManyWorlds→Many Worlds 2T conveyance and conveyances for other recently created asserter LLCs, that is the finding. It is not established yet.

4. Cascading transfers — not present

Only one post-issuance assignment (ManyWorlds, Inc. → Many Worlds 2T Innovations LLC, recorded 2026-04-15). No chain of LLC-to-LLC hops in <24 months. The "cascade" morphology is absent.

5. Pre-litigation transfer — present (strong)

The transfer to Many Worlds 2T Innovations LLC was recorded 2026-04-15. The same plaintiff filed Many Worlds 2T Innovations LLC v. Spotify Technology S.A., No. 2:26-cv-00303 (E.D. Tex.) on 2026-04-17 — two days later. The '603 patent is confirmed named in the later Many Worlds 2T Innovations LLC v. OpenAI OpCo, LLC, No. 2:26-cv-00774 (E.D. Tex., filed 2026-08-31), i.e. ~4.5 months after recordation — inside the 6-month window. Either way the chain was arranged immediately before assertion. (IPWatchdog, 2026-09-02, lists 12,299,603 among the five asserted patents: https://ipwatchdog.com/2026/09/02/openai-sued-over-ai-patents-covering-retrieval-personalization-content-generation/.) Caveat: I confirmed '603 in the OpenAI complaint; for the Spotify complaint I confirmed only certain family members (e.g., 8,655,829 and 8,676,742) in the accessible analysis, so the 2-day figure rests on family membership rather than a confirmed '603 count.

6. Bankruptcy fire-sale — not present

No bankruptcy proceeding for ManyWorlds, Inc. found, and the 2026-04-15 assignment is an ordinary recorded assignment, not a court-supervised §363 sale. The Kodak/Nortel/Polaroid-style fact pattern does not apply.

7. Privateering — unclear

The naming continuity ("ManyWorlds" → "Many Worlds 2T Innovations") and the fact that the original assignee's own principals (Flinn, Moneypenny) are the inventors suggests direct monetization by the founding team rather than a third-party NPE asserting on behalf of a still-operating company. Classic privateering requires an operating company that keeps selling into the market while an NPE sues its competitors; there is no evidence ManyWorlds, Inc. is a competitor of Spotify or OpenAI. Mark unclear — a revenue-share or retained-interest arrangement in the assignment would settle it, which is again a reel/frame question.

8. Defensive aggregator — not present

The chain terminates at an asserting LLC, not at RPX, AST, LOT, Unified, or OIN. The inverted (neutralized) outcome does not apply.

Adjacent pattern worth logging (not one of the eight): continuation-tuning to create assertable claims. This patent issued 2025-05-13 as a continuation claiming priority to a 2011-03-29 provisional via US 13/268,035 (2011-10-07) and US 13/295,414 (2011-11-14), with anticipated expiration 2031-10-07. Its sibling '388 (12,307,388) issued 2025-05-20. Fresh claims minted between 2024–2025 and asserted within ~5 months of the transfer is a monetization-shaped filing strategy. Also relevant: this family has a documented §101/Alice ex parte appeal history (Appeal 2013-002476 and 2018-006544, Flinn/Moneypenny, ManyWorlds assignee), so a §101 defense is foreseeable.

Verdict

NPE — high confidence.

The chain is short but decisive: ManyWorlds, Inc. (a small Texas software/consulting company that does not appear to sell any product reading on the claims) transferred the entire interest in US 12,299,603 to Many Worlds 2T Innovations LLC by assignment recorded 2026-04-15, and that assignee filed Spotify (No. 2:26-cv-00303) on 2026-04-17 and OpenAI (No. 2:26-cv-00774) on 2026-08-31 — the '603 patent being expressly named in the OpenAI complaint. That yields two independent strong signals, pre-litigation transfer (Signal 5) and shell-entity transfer-to-asserter (Signal 1), with the assignee's name literally encoding the accused "two-tower" architecture. The confidence is capped only by two unverified fields — the reel/frame and the recording correspondent — which I could not retrieve and have declined to fabricate; pulling those from https://assignment.uspto.gov/patent/index.html (search "12299603" or "Many Worlds 2T Innovations") is the one step that would either harden this to an airtight finding (a repeat recording attorney appearing across other asserter LLCs) or soften Signal 1 (if the LLC turns out to have a genuine operating footprint).

Search link for verification: https://assignment.uspto.gov/patent/index.html#/patent/search — query 12299603; and Google Patents legal events at https://patents.google.com/patent/US12299603B2/en#legalEvents.

I'll research US patent 12299603 and its prior art citations. Let me search authoritative sources.

The general results don't yet surface the citation record for this specific patent. Let me drill into the patent's own citation data.

Let me try alternative sources (Justia, FreePatentsOnline) that mirror USPTO citation records.

Summary of what I found — and an important limitation

I ran multiple searches across Google Patents, USPTO-mirroring aggregators, and general web sources for the specific number US 12,299,603 / US12299603B2. I could confirm the patent's identity, but I could not retrieve the patent's actual "References Cited" (prior-art citation) list, and I will not fabricate it. Details below.

1. Patent identification (confirmed)

These fields come directly from the authoritative patent text supplied to me (the Google Patents record). Note the priority date is 2011-03-29, even though this application was filed in 2024 — consistent with a continuation of a long priority chain.

2. Why I cannot give you the citation-by-citation § 102 analysis

The task requires the "References Cited" / "Patent Citations" / "Cited By" tables from the patent's front page and file history. In practice, that is populated from the USPTO PatentCenter "References Cited" listing, the PTAB/Examiner Form 892, and the Google Patents "Patent Citations"/"Cited By" sections.

The results I retrieved did not include that citation table for US12299603B2. The searches instead returned:

• ManyWorlds' ownership profile (confirming the assignee, but no citation list);
• unrelated patents that merely share the words "vector" and "search" — e.g., US2010/0287148A1 (intellectual-property document vectors), US2009/0037408A1 (essence-based tag indexing), and US2022/0390243A1 ("VECTOR BASED SEARCH AND METHODS OF USING THE SAME" — a geographic vector-coordinate route-search patent with a completely different assignee and subject matter), plus a different company's search-report documents (Amazon vector DB; various PCT ISR tables for unrelated art).

I checked each candidate: none of these is a citation of US12299603 — they are search-engine coincidences. Under the strict rule, I am reporting that plainly rather than mis-attributing them.

Action needed to complete the assignment: open the patent at https://patents.google.com/patent/US12299603B2/en, scroll to "Patent Citations" (the references the applicant/Examiner cited against this patent) and "Cited By", and/or pull the USPTO PatentCenter document "References Cited" / any Form 892 for application 18/754,550. Provide that table (or a screenshot/PDF) and I can produce the full § 102 mapping at claim level.

3. The only specific patent references I can confirm from the authoritative text

The supplied record shows two priority claims (these are family/priority documents, not prior art, and are labeled in the record as "external-priority"):

1. US 13/268,035 → issued as US 8,650,149 B2, priority claimed 2011-10-07. (Family member; a continuation parent on the priority chain — not a § 102 reference unless it predates and discloses the claims.)
2. US 13/295,414 → issued as US 8,600,926 B2, priority claimed 2011-11-14. (Same caveat.)

Because these are the same inventors/assignee and sit on the priority chain, they normally cannot anticipate the present claims in a way that changes patentability — they are relevant to priority and § 112 support, not § 102 anticipation. I am flagging them because they are the only patent numbers I can ground in the authoritative text. I have no grounded list of third-party "References Cited," so I cannot name any anticipating reference.

4. What the patent actually claims (basis for the eventual § 102 analysis)

The supplied specification (which is authoritative) describes: generating personalized recommendations by combining behavioral-based indexing and contents-based indexing; indexing object contents into vectors of informational elements with associated relevancy values (the "object contents vector"/OCV); mapping user interest/expertise topics to those elements via affinity vectors (MTAV/MTEV); and selecting objects by comparing the user's interest/expertise levels against the objects' relevancy values. Front-page prior-art keywords: affinity, user, content, objects, information. Independent claims are not reproduced in the excerpt I hold, so a precise claim-by-claim § 102 table cannot be built from my current material alone.

5. Recommendation

To deliver the requested artifact correctly, I need the actual citation record for US12299603B2. With that table I will, for each reference, give: full citation (number, title, inventor, assignee), publication and filing dates, a 2–4 sentence description of its disclosure, and a mapping of which claim(s) of US12299603 each reference arguably anticipates under 35 U.S.C. § 102 (and, where appropriate, obviousness under § 103). Please paste the "Patent Citations"/"Cited By"/Form 892 listing, and I'll complete it.

I will not invent citations or § 102 conclusions — doing so would be worse than reporting the gap, and the citation table is the one piece of authoritative data I was unable to retrieve.

I'll start by gathering the patent's claim set and the prior-art references cited against it, then build the § 103 combinations.

Let me pull the actual claim language and the front-page references.

Obviousness Analysis — U.S. Patent No. 12,299,603 B2 (“Vector-Based Search Method and System”)

Prepared under 35 U.S.C. § 103. Identifiers are read literally as they appear in the record.

0. Scope, posture, and explicit gaps in what I could verify

Before the analysis, three candor items — they materially affect how much weight you should put on the mapping below.

(a) The claim set was not in my page capture. The Google Patents capture supplied to me contains the bibliographic data, the classification tree, the “Definitions” extract, and the detailed description (through Table 1). It is truncated before the Similar Documents / References Cited / Cited By tables and does not include the claims. I therefore cannot quote or chart the issued claims.

(b) The “Prior Art” section of the page was likewise truncated. What I can read from that section is:

So the analysis below is grounded in (i) the examiner’s own field-of-search keywords, (ii) the specification’s admitted state of the art, and (iii) references in the pre-2011 information-retrieval/recommender literature that I can identify with confidence. If you paste the front-page reference list and the claim set, I will convert this into a limitation-by-limitation chart with pin cites.

(c) A separate dataset flags live litigation. A litigation tracker surfaced in my search lists “Many Worlds 2T Innovations LLC v. OpenAI OpCo, LLC” (E.D. Tex. 2:26-cv-00774, filed 2026-08-31), asserting — among others — “12,299,603” (same digit string as 12299603) against ChatGPT Memory/Search/Deep Research/Projects and OpenAI Vector Stores/File Search. That tells you how the patentee is reading the claims: onto modern vector-store retrieval. That framing is exactly what makes the § 103 exposure acute, because generic vector retrieval is 40+ years old.

1. Threshold issues that dominate the § 103 analysis

1.1 Which § 103 governs — and why it may not be the one you expect

The § 103 analysis is gated by the effective filing date of each claim.

• If every claim of the ’603 is supported by the 2011 disclosure (13/268,035 and/or 13/295,414), the effective filing date is on or about 2011-03-29 / 2011-10-07, and pre-AIA § 103(a) applies. Under pre-AIA § 102(b), references published more than one year before the earliest U.S. filing date to which benefit is claimed — i.e., roughly before 2010-03-29 (or 2010-10-07) — are statutory-bar art.
• If any claim contains a limitation not supported by the 2011 parents — e.g., “embedding,” “neural network,” “vector database,” “large language model,” “approximate nearest neighbor” — then under AIA § 3(n)(1) that claim gets a 2024 effective filing date, the AIA §§ 102/103 apply, and the prior-art universe explodes to include 2011–2024 art (word2vec 2013; FAISS/HNSW ANN indices 2017; transformer/BERT retrieval 2018–2019; dual-encoder retrieval; commercial vector stores).

Recommended first step for any challenger: side-by-side the issued claims against the 13/268,035 and 13/295,414 disclosures. A continuation filed 2024-06-26 in a 2011 family is a classic place where a limitation has drifted beyond the original disclosure. Note also that the ’603 issued only ~10.5 months after filing — consistent with a “bye” continuation off allowed parent claims; the file wrapper will show what was actually examined and what was cited.

1.2 The specification contains powerful § 103 admissions

The “Definitions” extract on the page is, functionally, a stipulation of the prior art:

“Existing computer-implemented recommender systems can provide personalized recommendations basis expected interests of recommendation recipients derived from behavioral history…”
“Existing search engines can index the contents of documents and retrieve information based on search terms…”
“these approaches operating independently are inadequate in some cases for delivering the most useful information…”

Under KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398 (2007), and MPEP § 2144.03, a specification’s identification of two known, inadequate approaches in the same field is close to a dispositive statement of: (i) what the claim elements are, and (ii) why a POSITA would combine them. The patent’s own “Summary” then says the invention is “generating personalized recommendations based, at least in part, on combining behavioral-based indexing and contents-based indexing.” That is the whole alleged invention, stated as a combination of known elements.

1.3 Level of ordinary skill

A POSITA circa March 2011 would be a software engineer/scientist with a BS (or MS) in CS and 2–3 years’ experience in information retrieval and/or recommender systems, familiar with: the vector-space model and TF-IDF; LSI/LSA; collaborative and content-based filtering; clickstream/usage-log mining; and web application APIs. No specialized “new” skill is required by the ’603’s disclosed techniques.

2. The limitation set a § 103 challenge must meet

Since I lack the literal claims, here is the limitation set that the specification supports and that a challenger should expect to attack. Treat labels A–H as a reconstruction, to be conformed to the issued claims.

Dependent-claim-level features to expect: MTAV built from subscribes/accesses/saves/tags; OCV-derived CTAVs (topics inferred statistically from OCVs); standardized tags for portable MTAVs across originating systems; MMAV (user-to-user) from standardized MTAVs; cross-system contextual correspondence; score harmonization across sources.

3. The prior-art universe (all pre-2011, so all clearly available)

Group I — Content indexing into weighted vectors (maps to B, D)

1. Luhn, A Statistical Approach to Mechanized Encoding and Searching of Literary Information, IBM J. Res. Dev. (1957) — term frequency weighting.
2. Salton & McGill, Introduction to Modern Information Retrieval (McGraw-Hill 1983) — the vector-space model: term vectors with weights, query vectors, cosine similarity. This is the direct ancestor of the “object contents vector.”
3. Salton & Buckley, Term-weighting approaches in automatic text retrieval, Info. Processing & Mgmt. (1988) — TF-IDF.
4. Deerwester, Dumais, Furnas, Landauer & Harshman, “Indexing by Latent Semantic Analysis,” JASIS 41(6):391–407 (1990) — term–document matrix reduced to a concept/topic space; i.e., deriving topics statistically from content vectors and computing object-to-topic affinities. Directly maps to the “OCV-derived CTAV” dependent claims.
5. Rocchio, Relevance feedback in information retrieval (1971) — query modification from feedback vectors (anticipates “supplemental” result expansion).
6. Hofmann, Probabilistic Latent Semantic Indexing, SIGIR 1999; Blei, Ng & Jordan, Latent Dirichlet Allocation, JMLR 2003 — statistical topic inference from content vectors.

Group II — Behavioral indexing / personalized retrieval (maps to C, E, F)
7. US 5,754,939 (Herz et al.) — user profile vectors constructed from word frequencies in articles the user read; object/target “profile interest summaries”; and, critically, search profiles formed by combining the user profile with the query and compared against object profiles. This is an OCV + MTAV + search-time personalization reference, all in one.
8. US 5,835,087 (Herz et al.) — companion reference for object profile generation.
9. US 6,078,916 (Culliss, Infoseek) — ranking retrieval results using observed user selection behavior (frequency of selection), i.e., behavioral data folded back into search ranking.
10. Direct Hit Technologies patents (e.g., US 6,006,222) — “click popularity” relevance scoring: behavioral data over result sets used to re-rank content-relevant results.
11. US 6,981,040 B1 (Konig et al.) — “Automatic, personalized online information and product services”: a learning machine continually updated from the user’s actions and similar users’ actions; content-based user profile vectors compared to document vectors.
12. Joachims, Optimizing Search Engines Using Clickthrough Data, KDD 2002 — clickthrough as implicit relevance signal for ranking.

Group III — Item-to-item / behavioral object-to-object affinities (maps to C, E)
13. US 6,912,505 (Linden, Smith & York, Amazon.com) — “Use of product viewing histories of users to identify related products”; US 6,317,722 (Jacob et al.) — cart-based recommendations. These generate item-to-item similarity from co-access/co-purchase histories — literally the CCAV of the ’603.
14. Linden, Smith & York, Amazon.com Recommendations: Item-to-Item Collaborative Filtering, IEEE Internet Computing 7(1) (2003).
15. Sarwar, Karypis, Konstan & Riedl, Item-Based Collaborative Filtering Recommendation Algorithms, WWW10 (2001).
16. Kleinberg, Authoritative Sources in a Hyperlinked Environment (1998/1999); Page et al., PageRank — structural/behavioral link affinities.

Group IV — Hybrid content + behavioral recommenders (the combination itself)
17. Balabanović & Shoham, “Fab: Content-Based, Collaborative Recommendation,” Comm. ACM 40(3):66–72 (1997). Fab maintains both (i) content vectors (TF-IDF term vectors of web pages, cosine-matched to a user profile) and (ii) collaborative user-rating vectors, and merges the two recommendation streams at a selection agent into one ranked list, with the user profile updated from feedback. This is the ’603’s claimed combination, in a 1997 paper.
18. Pazzani, A Framework for Collaborative, Content-Based and Demographic Filtering, AI Review (1999).
19. Smyth & Cotter, A personalised TV listings service for the digital TV age, Knowledge-Based Systems (2000) — hybrid content-vector + collaborative recommendation, with case-based profile updating.
20. Lieberman, Letizia: An Agent That Assists Web Browsing, IJCAI 1995; Pazzani et al., Syskill & Webert (1996) — implicit browsing behavior → content profile vectors used to recommend/expand navigation.
21. Mobasher, Cooley & Srivastava, Automatic Personalization Based on Web Usage Mining, CACM 43(8):142–151 (2000) (and Mobasher et al., Effective personalization based on association rule discovery from web usage data, WIDM 2001) — aggregate usage logs into transaction vectors, cluster into user profiles, and map usage clusters to content clusters to drive presentation.

Group V — Interoperability/portability (maps to the standardized-tag / cross-system claims)
22. Dublin Core (1995); W3C PICS (1996); RDF (1999); FOAF (2000); RSS/Atom (2000s) — standardized, portable topic/metadata vocabularies enabling profile reuse and cross-system comparison.

Caveat on numbers: I am confident about US 5,754,939, US 5,835,087, US 6,912,505, US 6,317,722, US 6,981,040, and US 6,078,916. For Direct Hit’s patents and a few others I am reasonably but not fully certain of the exact numbers; verify each against the front page before relying on it in a petition or office action.

4. The obviousness grounds

Why these are combinations, not merely juxtapositions

Under KSR, the challenger should articulate at least one of the following, as applicable to each ground:

1. Design need / market pressure. The ’603’s own background states that content-based and behavioral approaches “operating independently are inadequate.” Fab (1997) says the same. Where the specification and the art identify the same deficiency, the motivation is supplied by the record. KSR, 550 U.S. at 421.
2. Known technique improving analogous devices in the same way. Vector-space matching, latent topic reduction, item-to-item co-occurrence, and profile-based re-ranking were each routinely applied to improve retrieval/serving performance. Id. at 417.
3. Finite number of identified, predictable solutions. Hybridization of content and collaborative filtering was the recognized, enumerated design space (Fab; Pazzani 1999; Smyth & Cotter 2000). Combining two similarity scores (normalization, weighted sum, or source harmonization) is arithmetic, and produces no more than predictable results. Id. at 416–17.
4. Express teaching, suggestion, or motivation in the references themselves. Herz’s “search profile” and Fab’s merged selection agent are explicit combination teachings.

5. Anticipated rebuttals and how they fare

6. Collateral vulnerabilities worth flagging

1. Obviousness-type double patenting. The justia results surfaced sibling continuations sharing verbatim MTAV/MTEV text (“GENERATIVE INTERACTIVE VIDEO METHOD AND SYSTEM,” US 2024/0119321; “NEURAL NETWORK TRAINING FOR VIDEO GENERATION,” US 2025/0292125; and asserted patents US 10,699,202 and US 12,307,388). A family this dense invites ODP challenges, and a terminal disclaimer in the chain can raise its own issues.
2. § 112(a) enablement/written description for any claim element not described in the 2011 parents.
3. § 102(a)(2)/102(e) art in the family itself. The large ManyWorlds family may contain earlier-filed, later-published applications usable as § 102(e) art against specific claims.
4. The examiner’s search field (as reflected in the page’s prior-art keywords — affinity, user, content, objects, information) is broad and generic, which typically signals an art-limited but not conceptually-novel claim.

7. What I need from you to finalize

1. The issued claim set of US 12,299,603 B2 (all claims, with dependency).
2. The full “Prior Art” section as it appears on your copy of the page — specifically Similar Documents, References Cited, Cited By, and Cited By Examiner — which were truncated in my capture. The Similar Documents rows with priority dates before 2010-03-29 are your highest-value targets; the Cited By Examiner rows tell you what the examiner already considered (and thus what you must differentiate).
3. The specifications of US 8,650,149 and US 8,600,926 (the 2011 parents), to test the § 3(n)(1) priority / AIA-vs-pre-AIA question per claim.
4. The ’603 file wrapper (IDS, office actions, interviews, reasons for allowance) — relevant both to what was considered and to any § 325(d)/estoppel dynamics in an IPR.

With items 1–2, I can produce a claim-chart with element-by-element pin cites to Salton/Deerwester, Herz, Fab, Amazon, Culliss, and Mobasher, plus a motivation-to-combine section tailored to each independent claim.