Software Patents After Alice
58% of software patents get approved. 88.6% of Section 101 appeals fail. The difference is how you describe your invention.
In 2014, the Supreme Court decided Alice Corp. v. CLS Bank International and changed the test for software patent eligibility. A decade of Federal Circuit case law has clarified what survives. This guide covers what gets approved, how attorneys navigate prosecution, and what developers should document before filing.
Last updated: June 2026. This page is informational only and not legal advice. Consult a patent attorney for your specific situation.
Recent developments (June 2026)
- AGI SureTrack v. Farmers Edge: Federal Circuit affirms §101 ineligibility on five farm-data patents (June 2, 2026, precedential). Sensors plus GPS plus implement-profile lookup, ruled abstract on generic hardware.
- AI patent risk: 2026 firm consensus: Synthesis of AmLaw 50 IP-group publications on AI-drafted patent risk across §101, §102, §112, and inventorship.
The Test
The Alice decision created a two-step filter for patent eligibility under 35 U.S.C. Section 101.
If you could describe your invention as "doing [familiar thing] but on a computer," it is probably directed to an abstract idea. Intermediated settlement, organizing data, filtering content, calculating a price. These are abstract ideas.
If NO → Patent eligible. Stop here.
If the claim is directed to an abstract idea, does it include an "inventive concept" that transforms it? This means more than running the idea on a generic computer. The claim must show a specific technical improvement or an unconventional arrangement of components.
If YES → Patent eligible.
Hear the two-step test argued at the Supreme Court
From the oral argument in Alice Corp. v. CLS Bank International, argued March 31, 2014. Public-domain audio, U.S. Supreme Court.
"...coming up with an idea and then say, use a computer, is not sufficient."Mr. Phillips, for Alice Corp.
"...the claim has to recite something significantly more, something significantly more than the abstract idea itself."Mr. Perry, for CLS Bank
Full argument: supremecourt.gov audio · official transcript
The Numbers
Software patents get through at lower rates than mechanical patents. But the approval rate is far from zero.
Methodology note for these four numbers
Sophisticated readers will want to know what each headline number is measuring. The four stats above are not directly comparable; they use different denominators, art-unit scopes, and time periods.
- 58% software allowance rate. Composite aggregate from multiple PatentBots-derived and practitioner-reported series, 2024. Not the same as PatentBots' published three-year grant rate by USPTO technology center, which reports figures including ~80% for TC2100 (Computer Architecture and Software), ~84% for TC2400 (Networking/Cybersecurity), and ~74% for TC3600 (where many software business-method patents land). "58%" here is the §101-tilted aggregate practitioners typically encounter for the kinds of software claims most affected by Alice/Mayo; treat it as an order-of-magnitude indicator, not a definitive measurement.
- 74% mechanical allowance rate. PatentBots three-year grant rate for mechanical art units (2024 reporting). The 16-point gap between this and the software number is the meaningful comparison; both are measured the same way.
- 77% AI office actions with §101 rejection. Juristat, 2024, for art units associated with AI-tilted technology centers. Counts the share of office actions, not the share of applications that ultimately face §101 rejection (some applications get multiple office actions and may face §101 in only one). The application-level rate is lower; the office-action-level rate is the one that matters for response-strategy planning.
- 88.6% PTAB affirmance rate on §101 rejections. PatentDocs (2024) reporting on ex parte appeals where the §101 rejection was at issue: 692 appeals, 613 affirmances. Corroborated by MBHB summaries showing §101 affirmance rates of 87-91% across 2021-2024. This is the best-sourced number on this page.
For all four figures, the practical implication is the same: §101 is the high-stakes issue at examiner level, and PTAB appeal is a low-probability path to reversal. Methodology heterogeneity does not change that conclusion.
2026 update: the §101 rejection rate is climbing back
Dennis Crouch published corrected long-form §101 examiner data in Revisiting Sixteen Years of §101 After a Data Correction (Patently-O, May 27, 2026). Two data errors fixed in the original dataset: the OCR step had been silently failing on the longest office actions, and the bulk USPTO dataset had dropped a substantial number of rejections that were reconstructed via the USPTO API.
The corrected series is nearly identical to the original through early 2019, then diverges, with the gap widening through 2024 and 2025. Two findings from the corrected data:
- The §101 rejection rate reached roughly 15.5% by mid-2025, within striking distance of the pre-2019-PEG peak.
- The post-2019 rebound was substantially steeper than originally reported. The Patent Examination Guidance period bent the trajectory upward more than the original dataset showed.
- The small decline observed under the Stewart-Squires USPTO leadership is real but modest against the larger post-2019 trajectory.
Implication for the headline allowance rate: the 58% software allowance rate cited above remains the best available aggregate figure, but the §101 rejection component within that aggregate is climbing. Practitioners should expect the eligibility hurdle to keep getting harder, not easier, regardless of the broader allowance trend.
Sources: USPTO examiner office-action data via Patently-O custom classifier (Crouch, May 27, 2026). The headline 15.5% figure is the verified mid-2025 data point; earlier-year numbers and exact Stewart-Squires-era decline are paywalled.
What Survives Alice
These Federal Circuit cases established categories of software patents that survive the Alice test.
| Case | Year | What It Was | Why It Survived |
|---|---|---|---|
| Enfish v. Microsoft | 2016 | Self-referential database | Improved how the computer itself stored and retrieved data |
| McRO v. Bandai | 2016 | Automated lip-sync rules | Replaced subjective human judgment with specific, rule-based automation |
| BASCOM v. AT&T | 2016 | Internet content filter | Unconventional arrangement of known components at a specific network location |
| Finjan v. Blue Coat | 2018 | Behavior-based virus scan | New technical approach: analyzing code behavior rather than matching signatures |
| Core Wireless v. LG | 2018 | Small-screen UI improvement | Solved a specific technical problem (limited screen size) with a specific solution |
| Berkheimer v. HP | 2018 | Document processing | Shifted burden: examiner must prove with evidence that elements are conventional |
The Pattern
- Improves the computer itself, not just uses it (Enfish)
- Replaces human judgment with a specific technical rule (McRO)
- Unconventional arrangement of components producing new capability (BASCOM)
Hear it argued: when software improves the computer itself
From the oral argument in Enfish, LLC v. Microsoft Corporation, argued February 5, 2016 at the Federal Circuit. Public-domain court recording (via CourtListener).
Full argument: CourtListener recording
Component Count Matters
Claims combining components from different technical areas survive at different rates.
What Fails Alice
| Case | Year | What It Was | Why It Failed |
|---|---|---|---|
| Alice v. CLS Bank | 2014 | Intermediated settlement | Business method on generic computer. No improvement to computer functionality. |
| Electric Power Group | 2016 | Power grid data collection | Collecting and analyzing information is abstract. No unconventional approach. |
| IV v. Symantec | 2016 | Email filtering | Organizing data into categories is a fundamental practice. |
| American Axle | 2020 | Vibration attenuation | Applied a law of nature without specifying how. Too abstract. |
| Recentive Analytics v. Fox | 2025 | ML-based scheduling | Applied conventional ML to new data domain without improving ML itself. |
2025 Warning: AI/ML Patents
Recentive Analytics v. Fox Corp. (Federal Circuit, April 2025) is the first major ruling on ML patent eligibility. The court held that applying known machine learning to scheduling data was abstract. "We used AI to do X" is not patentable. "We designed a specific architecture that solves [technical problem] by [unconventional approach]" might be.
The Failure Pattern
- Automates a known human process on a generic computer (Alice)
- Applies a formula or algorithm without connecting it to a specific technical problem (American Axle)
- Uses established technology on new data without improving the technology (Recentive Analytics)
How Attorneys Navigate It
Frame the Technical Problem
The single most important prosecution strategy. Frame the invention around a technical problem, not a business outcome.
| Abstract Problem (Fails) | Technical Problem (Survives) |
|---|---|
| "Faster checkout process" | "Database queries timeout when cart table exceeds 1M rows under concurrent load" |
| "Better recommendations" | "Cold-start latency exceeds 500ms because collaborative filtering requires minimum interaction history" |
| "Improved scheduling" | "O(n2) scheduling algorithm fails above 10K events because pairwise constraint checking exhausts memory" |
| "Smarter data organization" | "Cross-table joins require full table scans without the proposed index structure, consuming 40% of database capacity" |
Layered Claim Strategy
Write claims at multiple levels of abstraction. Each level serves a different purpose.
Independent Claim
Technical improvement without implementation details. Captures competitors who solve the same problem differently.
Dependent Claim
Adds the general technique. Narrows scope but covers variations.
Specific Claim
Exact implementation. Fallback if broader claims face prior art.
Hardware Integration Language
How you describe software-hardware interaction affects Alice survival.
| Software-Only (Risky) | Hardware-Anchored (Stronger) |
|---|---|
| "A method for classifying data" | "A system comprising a processor and memory storing instructions that, when executed, classify sensor data received from an imaging device" |
| "An algorithm that optimizes" | "An apparatus comprising a network interface controller configured to optimize packet routing based on real-time congestion measurements" |
| "Outputting results on a device" | "Rendering a diagnostic visualization on a display coupled to an imaging sensor, mapping spatial coordinates to detected anomalies" |
Specification Depth Matters
The more technical detail in the specification, the harder it is to characterize as abstract. Detail also strengthens Section 112 (enablement), which reinforces Section 101.
| Detail Level | Alice Risk | Section 112 Risk |
|---|---|---|
| Minimal (1-2 pages) | HIGH | HIGH |
| Moderate (5-10 pages) | MEDIUM | MEDIUM |
| Comprehensive (20-30 pages) | LOW | LOW |
What Developers Should Document
Every item below maps to an Alice defense strategy. Document these before meeting your patent attorney.
| Document This | Why It Matters for Alice | Example |
|---|---|---|
| Specific technical problem | Anchors Step 1 away from abstract idea | "Sequential searches cause memory fragmentation, degrading retrieval by 40%" |
| Prior approaches + limitations | Establishes context, supports non-obviousness | "Existing caching caches credentials, not verification results" |
| Hardware interaction | Provides physical anchor for claims | "Receives raw pixel data from CMOS sensor. Renders on calibrated display." |
| Performance benchmarks | Proves technical improvement at Step 2 | "P99 latency: 180ms before, 2ms after. Memory: 4.2GB to 890MB." |
| Why not mentally feasible | Defeats "mental process" categorization | "Classifying 10M packets/sec using 47-dimensional vectors" |
| Architecture choices | Shows non-obvious design decisions | "B-tree with bloom filter over hash table because collisions degrade linearly above 1M" |
| Alternatives considered | Broadens written description | "Evaluated LRU cache (cold-start problem), round-robin (ignores load variance)" |
What Weakens Your Position
| Avoid This | Why It Hurts | Document Instead |
|---|---|---|
| Business outcomes only | "Improved revenue" is not a technical improvement | Pair business benefit with technical mechanism |
| Generic computer language | "Using a computer to process data" = abstract | Name specific components and their functions |
| Unsupported broad claims | "Applicable to all industries" without examples | Describe 2-3 concrete embodiments |
| Implementation without problem | "We built X" misses problem framing | Lead with the technical limitation you overcame |
Pre-Filing Checklist
Use these 10 items before meeting your patent attorney.
Is Your Patent at Risk?
Frequently Asked Questions
Are software patents dead after Alice?
No. Software allowance rates are approximately 58%. The Alice ruling eliminated certain types of claims (business methods on generic computers) but left room for genuine technical improvements. Every major Federal Circuit case since 2016 has identified categories of software that survive. The test is specific: does your invention improve the technology itself?
What about AI and machine learning inventions?
AI patents face higher scrutiny. About 77% of office actions in AI-related technology centers include Section 101 rejections. Recentive Analytics v. Fox (April 2025) held that generic ML applied to a new data domain is abstract. To survive, AI claims must show a specific architectural improvement or a specific technical problem solved in an unconventional way. The USPTO's 2024 examples (47-49) illustrate what works.
Do I need hardware to get a software patent?
Not strictly. Enfish survived with a pure database improvement and no hardware claims. But hardware anchoring provides an additional defense layer. Patents that tie algorithms to specific physical devices (sensors, displays, controllers) have stronger Section 101 positioning than those relying on generic references to "an electronic device."
What happens if I get a Section 101 rejection?
Most software patents face at least one. Common responses: point to specific technical improvements (cite Enfish or McRO), emphasize unconventional component arrangement (cite BASCOM), or challenge the examiner's conventionality assertion (cite Berkheimer, which requires examiners to provide evidence). Technical benchmarks and inventor declarations strengthen all responses. PTAB affirms 88.6% of 101 rejections on appeal, so the examiner-level response matters.
What is the Patent Eligibility Restoration Act?
Proposed federal legislation to reform 35 U.S.C. Section 101 by narrowing the "abstract idea" judicial exception. Would replace judicial exceptions with a statutory framework. Introduced multiple times in Congress. Not yet passed as of February 2026. If enacted, it would significantly change the Alice landscape by reducing examiner discretion.