Candidate arguments for counsel, ranked strongest-first — brainstorming inputs for counsel to evaluate, not a drafted response.
provide feedback to the AI model indicating the one or more portions to fine-tune the AI model (claim 1); automatically providing feedback to the AI model indicating one or more portions of the first representation that contributed to the approval or the rejection of the output (claim 7); provide feedback to the AI model indicating one or more reasons for flagging the output of the AI model as the anomaly (claim 14)
For counsel to weigh: each independent claim closes a loop that uses the anomaly-detection result to fine-tune or improve the AI model itself — a candidate technological improvement under MPEP § 2106.05(a) that integrates any recited exception into a practical application at Step 2A Prong 2 (Enfish/McRO line). The office action dismisses the feedback step for claims 7-13 as mere 'filed of use and technological environment' under MPEP 2106.05(h) (OA1 rejection 2(c)) and does not separately address, at Prong 2, that the feedback is directed to modifying the operation of the AI model rather than to a field of use. Counsel may argue that improving how a generative AI model performs by feeding back the specific portions of a representation that caused a rejection is an improvement to the functioning of the computer/model, not an abstract idea implemented on a generic computer. This lever targets the Prong 2 analysis directly, where a finding of integration ends the inquiry in the applicant's favor.
- —Claim 1: 'generate a detailed report outlining one or more reasons for blocking the transmission ... indicating one or more portions of the first representation that contributed to the match accuracy failing to satisfy the anomaly detection threshold; and provide feedback to the AI model indicating the one or more portions to fine-tune the AI model'
- —Claim 14: 'provide feedback to the AI model indicating one or more reasons for flagging the output of the AI model as the anomaly'
- —Office action, claims 7-13 Prong 2: characterizes 'automatically providing feedback to the AI model indicating one or more portions of the first representation that contributed to the approval or the rejection of the output' as 'filed of use and technological environment (see MPEP 2106.05(h))'
MPEP § 2106.05(a) — improvements to the functioning of a computer or to another technology; § 2106.04(d) — integration into a practical application (Step 2A Prong 2)
Risk The examiner will likely respond that 'fine-tune the AI model' is claimed at a high level of generality with no disclosed technical mechanism, and reassert it as linking the exception to a technological environment (2106.05(h)). Prosecution-history caution: framing the invention as an 'improvement to AI model functioning' characterizes the claims around the feedback loop and may narrow scope in the file wrapper as to what the claims cover; confirm the specification actually describes the fine-tuning mechanism before committing counsel to that characterization.
Likely examiner response◐ survives — moderate
For an improvement-to-technology theory to control at Step 2A Prong 2 under MPEP § 2106.05(a), the claimed improvement must be reflected in the claim language and be an improvement to the functioning of the computer or model itself — not merely a recited desired result. The examiner can respond that the hooks recite only 'provide feedback to the AI model indicating the one or more portions to fine-tune the AI model' (claim 1) / 'automatically providing feedback ... indicating one or more portions ... that contributed to the approval or the rejection' (claim 7) — i.e., outputting information — without reciting HOW the model is retrained or its parameters altered. On that reading the examiner can maintain the MPEP 2106.05(h) field-of-use / 2106.05(g) post-solution characterization (per OA1 rejection 2(c)), arguing the model is invoked as a generic tool that receives data and the loop is nominally recited (no fine-tuning algorithm claimed), so the exception is not integrated into a practical application.
How to adjust Strongest if counsel can point to specific specification passages describing HOW the feedback modifies model operation (a concrete retraining/parameter-adjustment mechanism) and tie the improvement to what the claim actually recites, framing it in the Enfish/McRO 'improvement to the model' register rather than 'better result.' If the claim recites only 'provide feedback' as an output with no operative fine-tuning step, consider amending to bring the model-modification mechanism into the claim so the Prong-2 improvement is anchored in claim scope — arguing pure integration on the current wording carries category-of-use risk.
input, into an AI model, the user prompt to cause the AI model to generate an output based on the user prompt; retrieve, based on the user prompt, the data over the time period (claim 1); block transmission of the report to the user device (claim 1)
For counsel to weigh: at Step 2B the office action concludes the additional elements are well-understood, routine, and conventional by citing only MPEP 2106.05(d)(II) categories ('receiving or transmitting data over a network,' 'electronic record keeping,' 'storing and retrieving information in memory') (OA1 rejections 1-2, Step 2B). Under Berkheimer and MPEP § 2106.05(d), a conventionality finding must be supported by one of the four evidentiary showings (a court holding, a citation to a publication, an official-notice-type statement, or an applicant admission). Counsel may press that the office action supplies no such support for treating the specific ordered combination — generating dual representations, comparing them for match accuracy against an anomaly-detection threshold, and gating transmission — as conventional, and that the generic categories cited do not reach that combination. This is an evidentiary defect in the rejection that the examiner must cure to sustain Step 2B.
- —Office action, claim 1 Step 2B: 'The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), "receiving or transmitting data over a network", "electronic record keeping," and "storing and retrieving information in memory")'
- —Office action provides no § 1.132 declaration, publication citation, or applicant admission supporting the WURC characterization of the claimed comparison-and-gating combination
MPEP § 2106.05(d) — Berkheimer; a WURC finding requires factual support (§ 1.132 declaration, publication, official notice, or admission)
Risk The examiner will likely respond that the elements are recited at a high level of generality and re-cite the 2106.05(d)(II) enumerated categories, and may add a supporting citation on the next action to cure the gap. Note this is an evidentiary/procedural lever, not a merits win — even if pressed successfully it shifts the burden rather than establishing eligibility.
Likely examiner response◐ survives — moderate
The examiner can respond that the elements flagged (receiving/transmitting data over a network, electronic recordkeeping, storing/retrieving information in memory) are exactly the categories MPEP § 2106.05(d)(II) enumerates as court-recognized well-understood, routine, conventional functions, and that citing those enumerated examples is itself the Berkheimer support MPEP § 2106.05(d) contemplates. The examiner can further invoke the settled rule that the abstract idea itself (the dual-representation generation, match-accuracy comparison, and threshold gating) cannot supply the inventive concept, so any novelty in that ordered arrangement is analyzed at Step 2A, not as 'significantly more' at 2B. Practically, this is a curable defect: the examiner can add a citation or an official-notice statement and re-issue, rather than withdraw the rejection.
How to adjust The ordered-combination angle is the durable part — press that the § 2106.05(d)(II) categories reach the individual generic functions but not the specific arrangement (generate dual representations → compare against an anomaly-detection threshold → gate transmission), for which the record shows no evidentiary support. Recognize this lever tends to force a better-supported rejection, not allowance; treat it as pressure to prompt an interview/amendment posture rather than a stand-alone win, given the examiner's high interview-to-allowance correlation.
based on the match accuracy failing to satisfy the anomaly detection threshold, block transmission of the report to the user device (claim 1); based on the comparison, flag the output of the AI model as an anomaly (claim 14)
For counsel to weigh: the office action addresses the additional elements one-by-one — labeling the AI-model step as generic-computer implementation (2106.05(f)), the receive/input/retrieve steps as pre-solution data gathering (2106.05(g)), and the block-transmission step as field of use (2106.05(h)) — but the record does not show an evaluation of the elements as an ORDERED COMBINATION as MPEP § 2106.05(I) requires. Counsel may argue that the combination of validating an AI output against retrieved source data, gating transmission on an anomaly-detection threshold, and looping the failure reasons back to the model is a specific, non-generic arrangement that must be assessed together at both Prong 2 and Step 2B. Characterizing anomaly-based blocking of an outbound report as mere 'field of use' arguably understates its role as an operative control on the system's output. This attacks the analytical completeness of the rejection.
- —Office action, claim 1: separately labels elements under '2106.05(f)', '2106.05(g)', and '2106.05(h)' and concludes 'the additional elements individually or in combination do not integrate the judicial exception into a practical application' without a combination-specific analysis
- —Claim 1: 'based on the match accuracy failing to satisfy the anomaly detection threshold, block transmission of the report to the user device'
- —Claim 14: 'based on the comparison, flag the output of the AI model as an anomaly'
MPEP § 2106.05(I) — additional elements must be considered both individually and as an ordered combination; § 2106.05(g)/(h) — insignificant extra-solution activity vs. field of use
Risk The examiner will likely respond that the boilerplate phrase 'individually or in combination' satisfies the ordered-combination requirement and reassert each element's category label. This is a completeness/procedural lever; standing alone it may not establish eligibility.
Likely examiner response◐ survives — moderate
The examiner can respond that MPEP § 2106.05 requires the elements be considered both individually and as a whole, and can assert that the office action did consider the combination and found it adds nothing beyond the sum of generic parts — i.e., that validating an output against retrieved data and gating transmission on a threshold is a conventional arrangement of the individually-identified generic elements. The examiner can also recharacterize 'block transmission of the report' as controlling a field of use (output delivery) rather than an operative technical control, reinforcing the 2106.05(h) label. Because this argument attacks analytical completeness rather than a substantive error, the examiner can cure it by adding an explicit ordered-combination sentence.
How to adjust This argument gains force only when yoked to Arguments 1 and 3 — the ordered combination matters if the individual pieces (AI generation, anomaly-threshold gating, feedback loop) are themselves shown to be non-generic or non-mental. Frame it as the completeness frame that carries the substantive points, not as a stand-alone procedural objection, and be ready for the examiner to simply supply the missing combination analysis.
input, into an AI model, the user prompt to cause the AI model to generate an output based on the user prompt (claim 1); apply a text filter to the output ... tokenize the plurality of text-based analytics to extract a first plurality of keywords (claim 3)
For counsel to weigh: the office action characterizes the generating and comparing steps as mental processes performable by a human 'using paper / pen / calculator' (OA1 rejection 1, Step 2A Prong 1). Counsel may argue that limitations requiring an AI model to generate the output, and requiring tokenization of text-based analytics to extract pluralities of keywords for category-based similarity comparison against thresholds (claims 3-6, 12-13, 19-20), are not steps that can practically be performed in the human mind, which under MPEP § 2106.04(a)(2)(III) removes them from the mental-processes grouping. The record ties the representations to the machine-generated AI output and to programmatic tokenization/keyword extraction, not to observation or judgment a person performs unaided. Whether the recited operations fall outside the mental-processes grouping is a Prong 1 question for the examiner to reassess.
- —Office action, claim 1 Prong 1(a): '"generate", "provide feedback ... indicating" ... encompasses generating data representation model, comparing data, and generating report for anomaly detection, which is based on observation, evaluation, judgement, and/or opinion, that could be performed by human using paper / pen / calculator'
- —Claim 3: 'apply a text filter to the output to generate a modified output ... tokenize the plurality of text-based analytics to extract a first plurality of keywords within the plurality of text-based analytics'
- —Claim 1: 'input, into an AI model, the user prompt to cause the AI model to generate an output based on the user prompt'
MPEP § 2106.04(a)(2)(III) — mental processes; limitations that cannot practically be performed in the human mind are not within the grouping
Risk The examiner will likely reply that the AI-model and tokenization steps were treated as additional elements (not the abstract idea itself) and that the representation/comparison logic remains a mental process at a high level of generality. Prosecution-history caution: arguing the operations are inherently machine-scale may be read as conceding the underlying comparison concept is otherwise abstract; frame carefully.
Likely examiner response◐ survives — moderate
The examiner can concede AI-model generation is not literally done in the head yet respond that the identified judicial exception is the comparing/anomaly-detection/keyword-similarity evaluation — a mental process of observation and judgment — and that the AI model and tokenization are recited at a high level of generality as tools that merely apply the exception (MPEP § 2106.05(f)), which does not remove the claim from the mental-processes grouping. The examiner can characterize tokenizing text and extracting keywords for category-based comparison as a parsing/matching step a person can perform with pen and paper, per the OA1 Prong-1 framing, so the recitation of a generic AI model does not defeat the mental-process finding.
How to adjust Separate the two sub-arguments: the AI-model-generates-the-output point is stronger (generation of a generative-model output is not practically performable in the mind), while the tokenization/keyword point is weaker because parsing/matching is more readily cast as a mental step. Anchor the argument in specific claim language and specification detail showing the operations are machine-bound (e.g., what the AI model does that a person cannot), and pair with Argument 1 so the Prong-1 win narrows the exception even if the comparison step remains characterized as mental.
A system comprising: a storage device; and one or more processors communicatively coupled to the storage device storing instructions thereon (claim 14)
For counsel to weigh: at Step 1 for claims 14-20 the office action states 'The claim recites a process' (OA1 rejection 3, Step 1), but claim 14 recites 'A system comprising: a storage device; and one or more processors,' i.e., a machine, not a process. The office action's Step 2A Prong 1 analysis for claims 14-20 also appears cut off mid-sentence (OA1 rejection 3 notes the text is truncated), so the full articulated basis for the rejection of claims 14-20 is not on the present record. Counsel may argue the examiner has not set out a complete prima facie eligibility rejection for these claims and has misidentified the statutory category, which bears on the correct additional-element/Prong-2 analysis for a machine claim. This is a defect in the rejection's articulation that the examiner should correct or complete before the rejection can be sustained against claims 14-20.
- —Claim 14: 'A system comprising: a storage device; and one or more processors communicatively coupled to the storage device storing instructions thereon'
- —Office action, claims 14-20 Step 1: 'The claim recites a process, which falls into one of the statutory categories.'
- —OA1 rejection 3: 'The available text of the office action for claims 14-20 is cut off mid-analysis at Step 2A Prong 1'
MPEP § 2106.03 — statutory categories (machine vs. process); § 2106 — a complete prima facie eligibility rejection must be articulated
Risk The examiner will likely treat the 'process' label as a harmless typographical error and supply a corrected, complete analysis in the next action that parallels claims 1-6/7-13, so this may only prompt a cleaned-up rejection rather than withdrawal. Verify against the full mailed office action whether the ¶9 text is genuinely incomplete or merely truncated in the provided excerpt before relying on the incompleteness point.
Likely examiner response⚠ fragile — the comeback likely defeats it
The examiner can treat the 'process' label for claim 14 as a harmless clerical error — claim 14 is a machine (storage device plus processors), but a machine claim can still be directed to a judicial exception under Alice, so the category label does not change the eligibility outcome; the examiner will simply correct 'process' to 'machine' and reach the same Step 2A/2B result. The truncated ¶9 analysis is a record/formatting matter the examiner cures by issuing a complete or corrected action, and the parallel structure of claims 14-20 to claims 1-13 lets the examiner map the same reasoning without substantive change.
How to adjust This is a procedural/articulation point, not a patentability lever — it forces a complete prima facie case for claims 14-20 (legitimate, since an incomplete rejection is not sustainable on the present record) but does not advance the merits and is trivially curable. Use it to require a full articulation and to preserve the record, but do not lead with it; the substantive Prong-2/Prong-1 arguments for 14-20 should track the machine-claim analysis once the category is corrected. Consider raising it in an interview rather than spending written-response weight on it.