Freedom-to-operate patent analysis, explained.

Freedom to operate explained: a patent you own does not give you the right to sell

One FTO update, from the product's facts to a signed conclusion

1. The product is written down, feature by featureMRW-217: antibody class, target, concentration, buffer, sugar, surfactant, viscosity, dosing, device, process. Each fact points to its document.
2. Searches find what is near the productBy target, formulation, dosing, device and process, in every launch market. Fictional counts: 1,860 documents in 412 families.
3. Each live claim is read against the productElement by element, with the deciding element named. This is where the European claim is caught, or missed.
4. The team tests a change while it can still changeA different buffer, a different device. No change comes free: one fix can bring in a different owner's claim.
5. A patent attorney concludes and signsDesign around, license, challenge, monitor or do nothing. The date of that decision matters later.

Step 3 is the center of this guide. The people, companies, patent numbers and figures in this example are fictional. The patent numbers are set past any range the offices have reached.

What trips up FTO newcomers: sister patents, unpublished filings, thin public lists and office searches

Claim, element, patent family, claim chart: four terms to start with

Fourteen terms are defined in the one-page summary.

An FTO search finds patents, a claim chart reads them, and only a patent attorney gives the opinion

In shortPeople use "FTO" for four different things with very different prices and very different weight. A search, sometimes called a patent clearance search, finds documents. A landscape shows who is active. A chart reads claims against one product. An opinion is a lawyer's reasoned conclusion. Know which one you are holding.

The confusion matters because each of the four can look complete. A report with fifty patents in it can feel like an answer, and it leaves the question open until someone has read the claims against your product. The grid below shows what each one does.

Even the word "freedom" oversells it. No FTO opinion promises that nobody will sue. It says that, on what could be found on that date, a lawyer's reasoned view is that the product does not infringe a valid claim, or that it may, and what to do about it. The date is part of the opinion.

Reading a patent claim element by element: one missing element and the claim does not read

In shortA claim reads on a product only if every element is there. US law calls each element material. So one number outside a range, or one ingredient that is absent, can decide the question. A literal "does not read" is still a first answer: a court can also ask about equivalents, and the file history can narrow or widen what is left.

Here is claim 1 of the guide's fictional European patent, split the way a chart splits it, against the fictional antibody. Ranges, units and temperatures are checked exactly as written.

cP (centipoise) is a unit of viscosity. mg/mL is milligrams per milliliter. mM (millimolar) is a unit of concentration. % w/v means grams per 100 mL.

Now change one fact. A sodium acetate buffer at 20 mM and pH 5.3 has no histidine, so element f fails and the whole claim stops reading literally. That is why a formulation team can move a patent problem with a change that looks small. It also shows why element e matters: the unmeasured number stays unclear until somebody runs the test, and running it may be cheap.

Why the sister patents in one family give three different answers

The fictional family has three members. The US office narrowed its claim to one antibody class during examination, and the Japanese and European claims kept their wording. The same product gets a different reading from each.

Beyond the literal words: equivalents, the file history and the description

The element-by-element reading comes from settled law. In Warner-Jenkinson (1997), the US Supreme Court wrote: "Each element contained in a patent claim is deemed material to defining the scope of the patented invention."7

The same case says what can reach past the words. Under the doctrine of equivalents, a product missing an element literally can still infringe if it has an equivalent of that element, tested element by element. And under prosecution history estoppel, what a patent owner gave up during examination to win the patent can limit that argument.7 So reading the file history can help you or hurt you.

Europe adds a point on the reading itself. In G 1/24 (June 18, 2025), the European Patent Office's Enlarged Board of Appeal held that the description and drawings "shall always be consulted" to interpret claims when assessing patentability, even where the claim looks clear. We read law firms' reports of the decision, not its full text.8

The decision concerns patentability at the EPO, and courts apply their own tests, so it does not settle every forum, though the same firms say it moves the EPO closer to national courts and the Unified Patent Court. A chart built from claim text alone remains a good first filter. It stops short of a reading a lawyer can stand behind.

These are the limits of any chart, whoever or whatever fills it in. A "does not read" means the words do not fit. The questions of equivalents, the file history, the description and whether the claim is valid at all go to the patent attorney.

Why biologics are hard to clear: six layers of claims, many owners and thick patent families

In shortA small-molecule drug often turns on one compound patent. A biologic sits under layers: sequence, target and epitope, format, use and dosing, formulation, and manufacture and device. Often different owners hold different layers. A search on the target name alone will miss the layers that matter most late in development.

Researchers Michael Frakes and Melissa Wasserman built a database of all 515 FDA-approved biologics and over 11,500 patents, and report that patent thickets (dense sets of overlapping patents on one product) and evergreening (later patents that extend protection) are more common for biologics than for small molecules, and that patents delay biosimilars more effectively than they delay generics.9

The most quoted count is older and comes from an advocacy group, I-MAK. Its 2018 review of the 12 top-grossing drugs of 2017 counted, on average, 125 patent applications and 71 granted patents per drug, and 247 applications on one antibody.10 Those numbers describe sponsors' own patents on their own products. Your FTO has to deal with everybody else's too.

The six layers, and what each one means for a team trying to clear a product

An epitope is the part of a target that an antibody recognizes, and binding residues are the amino acids involved. The layer list draws on a search vendor's public guide; we keep it because practitioners use much the same one.11 A senior IP counsel at a drug maker argued in Bloomberg Law in February 2026 that FTO "can no longer be treated as a narrow legal exercise conducted late in development", and that assessments must evaluate method-of-use, formulation, manufacturing and combination claims.12

Antibody patents themselves changed in 2023. In Amgen v. Sanofi, a unanimous Supreme Court held claims covering an entire class of antibodies defined by what they do were not enabled, meaning the patent did not teach skilled people to make and use the whole class. It wrote: "The more one claims, the more one must enable."13

That ruling is about validity, so it can matter to the attorney's view of a broad functional claim. It takes no claim off a chart. A claim stays on the chart until a court or an office has removed it, and an attorney decides how much weight a likely validity problem deserves.

The law behind an FTO: infringement, the clinical-trial safe harbor, willfulness and what changed by 2026

In shortMaking or selling what a live claim covers infringes it. Work reasonably related to an FDA filing is protected in the US, and launch is not. Courts can triple damages in egregious cases, judged by what you knew then. No statute requires an opinion. Since 2025 the European courts reach further, and US patent challenges are harder to start.

What counts as infringing. In the US, whoever without authority "makes, uses, offers to sell, or sells" a patented invention, or imports it, infringes.14 Importing a product made abroad by a patented process, or selling it here, also counts, which is why a process claim can matter to a product made elsewhere.14 Europe has a similar rule: under Article 64(2) of the European Patent Convention, a process claim extends to the product directly obtained by the process.15

What development is protected. A US safe harbor, section 271(e)(1), covers uses reasonably related to developing and submitting information to FDA. In Merck v. Integra (June 13, 2005), the Supreme Court held it reaches preclinical work (before human trials), "as long as there is a reasonable basis for believing that the experiments will produce" the kinds of information relevant to an FDA submission.16

Europe's version, the Bolar exemption in Article 10(6) of Directive 2001/83, has grown unevenly: an International Bar Association review describes Germany's courts as function-based, Poland's Supreme Court as restrictive and Italy's Court of Cassation as somewhere between.17 So commercial launch sits outside all of these, and in Europe some trial work may too, depending on the country. Plan the FTO for the launch date, and plan a separate look at trials in countries with narrow exemptions.

What happens if you are found to infringe. A US court "may increase the damages up to three times".18 In Halo (June 13, 2016), the Supreme Court said such punishment should "generally be reserved for egregious cases typified by willful misconduct", and that culpability "is generally measured against the knowledge of the actor at the time of the challenged conduct".19 Enhanced damages are for the district court's discretion, on a preponderance of the evidence.19

Section 298 says failing to obtain or present advice of counsel may not be used to prove willfulness or intent to induce.20

So two common statements are both wrong. "You must have an opinion or you will pay triple" overstates it: the law imposes no such duty. "Opinions no longer matter" understates it: an honest, dated record of what you looked at and what you concluded is what you can show about your state of mind, if you choose to waive privilege, the protection that keeps lawyer-client communications confidential. That choice belongs to counsel.

Status of the rules an FTO team watches, as of November 3, 2026

When a blocking patent expires: 20 years, term adjustment, extensions and SPCs, worked through

In shortStart with 20 years from filing. Add any US term adjustment, then any extension: up to five years in the US and in Europe, plus six months for pediatric work in Europe. Then subtract the launch date. A claim that dies before launch is off the list, and a claim that lives nine years after it is a different conversation.

Here are the sums, using the guide's fictional family. Each number is on the fictional record, so you can check the arithmetic.

The extra years a patent can gain after approval

United States. Section 156 restores some of the time lost to FDA review. The extension may not exceed five years, the total after approval may not exceed fourteen years, and only one patent may be extended for a given review period.31 So when a third party owns the blocking patent and has no approved product on it, it will usually have no extension. Whether that holds for a patent on your list is a fact to check.

Europe. A supplementary protection certificate, SPC, starts when the basic patent ends. Its length is the gap from patent filing to first EU authorization, minus five years, and it may not exceed five years; pediatric work can add six months.32 A gap of 12 years gives 12 minus 5, or 7, capped at 5.

Since 2019, makers of generics and biosimilars have an exception for making during the certificate for export, or for stockpiling shortly before it ends.33 Cooley notes in August 2026 that the unitary SPC framework is still evolving.21

Other launch markets. Japan, Canada, Australia and Switzerland have their own extension rules. Check each with local counsel and record what the office itself shows. Do not assume European dates carry over.

What an FTO search must cover: patent full text, classes, registers and the public lists that do not help

In shortSearch full text and classification codes in every launch market, collapse to families, then check status office by office. Pending applications count, because their claims can still change and can still issue. Public drug lists are a small part of the picture, and the first 18 months after filing cannot be searched at all.

The Purple Book. Since the Biological Product Patent Transparency Act of December 2020, a sponsor must send FDA the patent lists it has served on biosimilar applicants, and the lists appear publicly, starting June 2021.3

The lists come from the patent exchange in the biosimilar law, in which the sponsor names patents it believes could reasonably be asserted.4 That step is optional in practice: in Sandoz v. Amgen (June 12, 2017), the Supreme Court held the applicant's duty to share its application cannot be enforced by a federal injunction.34 So a product with no biosimilar applicant shows nothing, and a third party's patent will not show at all.

Classification codes. Patent offices sort patents into classes. Antibodies against cytokines sit in C07K 16/24, for example.35 A search that uses words alone misses patents that describe the same thing differently, so analysts combine words, classes and owners. They search sequences on a separate route too, because a sequence claim may carry no target name.11

Families. Collapse documents into families, and read at least the member for each launch market. That single discipline is what catches the fictional European claim.

Status. Each office has its own register. A family granted in one country may lapse, be opposed or be narrowed in another. The European Patent Office's register shows oppositions; each national office shows renewals; courts show litigation. Status is a date-stamped fact, and it goes stale quickly.

Pending applications. Publication is not grant, and a published claim often changes before it issues. It still counts. In the US, a pending claim gives no damages until grant, though section 154(d) gives a reasonable royalty from publication if the issued claim is substantially identical and the party had actual notice.30 A watch on pending claims is therefore a routine part of the job.

The blind period. US applications generally publish promptly after 18 months from the earliest date, unless the applicant certified it will not file abroad.1 International ones publish shortly after 18 months.2 Everything filed in the last year and a half is unseen. A good FTO states that limit and repeats the search as the launch approaches.

Seven FTO mistakes that let a blocking claim through: one family member, a target-only search, a stale profile

In shortThe misses below come from assumptions: one family member stands in for the rest, a keyword search stands in for the layers, an old profile stands in for the current product. No one publishes an FTO error rate, so measure your own on a past update.

Later patents really do arrive after launch. In the aflibercept dispute covered in section 10, Goodwin's Big Molecule Watch reports the owner filed a new complaint on June 17, 2025, on a patent issued that same day, after the biosimilar launched, and a claim construction hearing set for November 23, 2026.37,38

A complaint is an allegation. The biosimilar maker's August 2026 quarterly report records a July 7, 2026 order on some defenses and counterclaims in that case, which did not decide infringement.39 We found no infringement ruling as of November 3, 2026, and we did not read the court docket.

What a blocking patent can cost: litigation, a European injunction and a late formulation change

In shortA patent owner's median cost to enforce runs from about $600,000 to $3 million at the sizes AIPLA reports, as cited in a sponsored article, and our reading is that a biologic launch usually has more at stake than those brackets. In Europe a preliminary injunction, a court order that can stop sales before the final judgment, takes about four months when heard with both sides. And changing a formulation after the lock adds months to a program that loses hundreds of thousands of dollars of sales a day.

What the Unified Patent Court is doing to launch plans

Carpmaels & Ransford's count of the Unified Patent Court's record to January 21, 2026 gives the clearest public view, on what its authors call a relatively small sample. Patent owners won 38 of 75 decided infringement cases, or 51%. Applications for preliminary injunctions were granted 29 times and refused 25 times, a 54% success rate among those decided either way.

The authors separately count 10 applications made without notice to the defendant, 6 granted and 4 refused. Injunction proceedings heard with both sides took just over four months on average, and full infringement cases ran at a three-month rolling average of just over 15 months.41

Reach grew as well. The EU's top court held in BSH v Electrolux (February 25, 2025) that a court in the defendant's home member state keeps jurisdiction over infringement of a patent granted in another member state, even when the defendant attacks that patent's validity, though only the other state's courts can rule on the validity itself. For a third country, the court may rule on validity as a defense, without changing that country's register. The ruling keeps a stricter rule for states in the Lugano Convention, whose courts have exclusive jurisdiction over validity of their own patents.22

A UPC division in The Hague then granted a permanent injunction in case UPC_CFI_386/2024, against a Dutch defendant, covering every state where the patent was in force, including the UK and Switzerland. For Switzerland, a Lugano state, it asked whether national revocation was a serious possibility, and for the UK it ruled on validity between the two parties. We read a law firm's note of that decision.43 These cases are young, and appeal rulings may refine them.

What a late formulation change adds

Changing a buffer after a Phase 3 lock means new stability data and a comparability exercise. ICH Q5E, an international guideline FDA adopted in June 2005, sets the principles for showing that a biological product is comparable before and after a manufacturing change.44 A comparability study can stretch to a bridging study, which supplies nonclinical or clinical data, if the lab results leave questions.44 No public source prices that for buffers, so treat any figure as your own CMC team's.

Delay is easier to price. Tufts CSDD works from one dataset of 645 drugs and publishes two figures for sales lost per day of delay: about $500,000 in the journal article, and about $800,000 in its August 2024 white paper. Both give about $40,000 a day of direct Phase II and III cost.45,46

After a blocking claim: design around, license, challenge or monitor, and the buffer case that shows the limits

In shortTeams have four main moves. A design-around changes the product. A license buys permission. A challenge attacks the patent. Monitoring waits for an opposition or a lapse. Each has a cost and a clock, and one route got harder: the US office now institutes far fewer challenges. Counsel chooses.

Oppositions and revocations: how often patents fall in Europe

Oppositions run on a clock: nine months from the notice of grant in the European Patent Bulletin.15 A preliminary view from an opposition division is not an outcome, and a patent amended in opposition may no longer read, or may still read. So a team watching someone else's opposition keeps the claim on the list until the decision, and keeps a plan for either result.

US patent office reviews start far less often than they did in 2024

In an inter partes review, someone who does not own a patent asks the US patent office to cancel its claims as unpatentable, relying on earlier patents and publications.48 Institution is the office's decision to start the review.

You may also read of a fall from about 65% in October 2024 to about 37% in February 2026. That pair comes from IPWatchdog's report of a USPTO webinar chart.50 The 65% appears only there, with its base unstated, and we read no USPTO table that gives it.

The 37% is USPTO's figure for fiscal 2026 through February, before closing at 41%: 236 instituted of 640 decided, by petition, inter partes and post-grant reviews together. The 41% is the same count for the full fiscal year, to September 30, 2026: 379 of 931. By patent, the year closed at 44%. Bio/pharma petitions ran higher, at 55%, on a small count of 39 of 71.49

A rule proposed October 17, 2025 set out categorical bars on institution, meaning fixed grounds for refusing to start a review. Those include claims already upheld elsewhere and cases where parallel litigation is likely to decide validity first.24 OIRA, the White House review office, shows it received the final rule on July 22, 2026, with review still pending on November 3, 2026.25 Until the text is out, nobody can state its final terms. Any FTO plan that reads "we would challenge it at the Patent Office" deserves a fresh look from counsel.

What the Federal Circuit said about a buffer in March 2025

The clearest public example of a formulation question at the US appeals level is Federal Circuit appeal No. 2024-2351, decided March 14, 2025, between the patent owner, Regeneron, and the biosimilar maker, Amgen.23 It came on a preliminary injunction, so it decides what the owner had shown so far and no final question of infringement.

The patent's claim 1 lists four parts of a formulation separately, among them "a vascular endothelial growth factor (VEGF) antagonist" and "a buffer". According to the opinion, Amgen's product "does not contain a separate buffer component", because "the aflibercept itself provides sufficient buffering capacity to stabilize the formulation".23

The court applied its Becton line of cases: where a claim lists elements separately, the implication is that they are distinct components. Finding that the claims and the specification (the patent's description) reinforced that reading, it held there was "at least a substantial question of noninfringement", that the owner "has therefore not established a likelihood of success on the merits", and affirmed the denial of the injunction.23

Three lessons follow for anyone planning a change. Claim structure counts: how a claim lists its parts can decide a case. The specification counts: the court read the patent as a whole. And the ruling had limits: the opinion records that other applicants' versions did contain a separate buffer, and those disputes went differently at the preliminary stage. A preliminary ruling in either direction finds no one liable.

What AI changes in patent analysis, which rules reach it, and where a lawyer stays in charge

In shortSoftware can read far more claims and check ranges exactly. It can also make things up. Rules for drug makers mostly do not cover this work, and rules for lawyers do: verify what a tool produces, and protect confidences. We found no independent test of claim-to-product matching on a real FTO. Test any tool on your own past work.

Which AI rules reach FTO work, and which stop at FDA filings or manufacturing

Splitting claims against construing them: what software does and what the attorney keeps

The FTO trend: European patents go unitary, biosimilar fights grow, and nobody publishes the market size

In shortNo credible public figure sizes the FTO market. What can be measured is the ground under it: more European patents with unitary effect, a busier European court, and steady biosimilar litigation in the US. All of it pushes FTO from one review toward a running job.

A unitary patent is one right across the participating states, so a single action can reach them all. Cooley notes that medical and veterinary patents have higher unitary uptake than average, and that about two-thirds of newly granted European patents are not being opted out of the court, as of June 2026.21 More of the patents on your chart will be enforceable in one place.

In the US, Goodwin's Big Molecule Watch tracks biosimilar patent litigation case by case, and its 2025 review covers disputes over several molecules, with settlements fixing launch dates for several aflibercept biosimilars.37 Our reading: patent claims often shape the launch dates companies negotiate.

Teams are shifting too. The Bloomberg Law piece quoted earlier, by a senior IP counsel at a large drug maker, argues FTO should be ongoing and cross-functional.12 Vendors now sell claim charting and speed. We read no independent data on how many companies work this way, so take the direction as commentary. Market size: no public figure we could verify. What you can measure is your own: updates a year, days per update, claims read and claims missed.

Who runs an FTO update today: IP analyst, CMC lead, patent counsel and where 20 working days go

In shortAn analyst searches and charts, the CMC and clinical teams supply the facts about the product, and a patent attorney concludes. By hand, the slow parts are reading claims and matching them to documents. Teams use outside firms, national offices or an in-house group with a search platform, and most mix them.

The people in an FTO update, from IP analyst to head of patents

Outside firm, national office or in-house team: the ways FTO gets done and what each costs

Finland's patent office offers one published price point: a search at €2,670 in 2026, before 25.5% VAT, usually delivered in about three weeks, with about two days of an examiner's search time and a report of status and the ten most relevant documents, and no view on infringement.5 Law firm prices are rarely published. One law-practice software vendor's guide gives the tiers below.6 Treat those as a vendor's estimate: a biologic with many layers in many countries can run far above them.

FTO update 3 for MRW-217: a plain-words search, a European claim, an acetate buffer and signed conclusions

In shortThis is the example from the top of the page, inside the product. Five AI agents, each with one job, search, read, match, check status and rank. Elena Marsh, head of patents, reads the riskiest claim, tests a buffer change and signs her conclusions. Nothing on these screens is a legal opinion, and the report for counsel, called the pack, says so on every page.

1

A scope in plain words becomes six search sets

Jonah Pereira, the IP analyst, writes the scope: anything that could block MRW-217 in its launch markets. He then lists antibodies to its target, high-concentration formulations, dosing in ulcerative colitis, syringes and autoinjectors, and fed-batch manufacturing processes. The search builder turns that into six search sets with keywords and classification codes, among them one for owners to watch.

The run starts at 6:00 am and finishes at 7:44 am. Each query stays on the record with its count, so counsel can see what was and was not looked for.

2

From 1,860 documents to 23 claims for counsel

1,860 documents collapse to 412 families. The legal status checker keeps 168 families granted or pending in at least one of eight launch markets. The claim reader splits 1,214 independent claims into 5,930 elements. The element matcher finds 61 claims that touch a feature. 38 fail on a deciding element and stay on the record with that element named. The landscape summarizer ranks the other 23: 2 likely to read, 6 unclear, 15 do not read.

3

The riskiest claim, already charted

At 8:52 am Elena opens EP 8 912 448 B1, claim 1, from Strathmore Biologics: granted February 14, 2024, expiring September 22, 2038, 9.1 years after the US and EU launch. It is under opposition by Kerrow Pharma, with oral proceedings on November 18, 2026. The patent has unitary effect in the UPC states and national parts in Spain, the UK and Switzerland, which appear as dark chips in the markets row.

Five elements read, and element e stays unclear, which makes the claim "likely reads". The chart says what would decide it: viscosity at 25 °C, from CMC.

4

Viscosity was measured at 20 °C, so the element stays unclear

Formulation report DR-217-07, Table 4, gives 18.6 cP at 20 °C and says 25 °C was not measured. Viscosity falls as temperature rises, so the matcher leans toward "reads", and it still will not call the element read. Not measured means unclear. Each verdict opens at its document.

5

Elena switches the buffer to acetate, and two claims stop reading

Dr. Aisha Oduya's team has a development batch, DB-217-A12, in sodium acetate 20 mM at pH 5.3, six months on stability: aggregates 0.9% against 0.8% for histidine. Elena sets the buffer to acetate. Element f flips to does not read, and every chart is read again. Two claims stop reading and one starts. The European claim drops to 9th of 23. The new top claim is Hollinmere Immunology's US 19,739,152 B2, with a risk score of 34, unclear because the epitope has not been mapped amino acid by amino acid.

This is a literal reading. Whether an acetate buffer could count as an equivalent of the claimed histidine, and what the file history says about that, is a question for Elena and outside counsel.

6

No design-around is free

Acetate brings Kerrow Pharma's pending US 2026/0911842 A1 into view. Its claim 1 reads on the acetate formulation with a risk score of 30, and its claims may narrow before grant. Adding an autoinjector brings Ensley Devices' US 19,452,880 B2 in with a risk score of 65. The cost line for each scenario comes from CMC: acetate is one comparability study, with the lock kept if decided by November 6; acetate with an autoinjector adds device verification and a human-factors study, about nine months.

7

Counsel concludes, and the agents never do

Three gaps decide eight claims: viscosity at 25 °C, residue-level epitope mapping with a competition assay, and the composition of a supplier's feed. The request is drafted and waits for Elena: nothing goes to CMC until she approves it. She records her conclusions: design around for the European and Japanese formulation claims, further data needed for the US method claim. Five claims stay open for the next update. She signs. The signature carries her name, role and date, and the analysis stays attached as the basis, marked as no legal opinion.

Steps 2 and 5 in numbers: the cut from documents to claims, and what the acetate switch did to the list.

How one European claim ends up with a risk score of 40

The score sorts counsel's reading list, and it is a plain formula with weights in Settings. It predicts no outcome and says nothing about validity.

For EP 8 912 448 claim 1: a "likely reads" verdict weighs 0.85. The markets where it is in force hold 26% of the launch forecast (UPC states 17, UK 5, Spain 3, Switzerland 1), and the formula turns that into 0.4 + 0.6 × 0.26 = 0.556. It has at least five years after launch, so its time factor is 1. Granted and opposed weighs 0.85. So 100 × 0.85 × 0.556 × 1 × 0.85 = 40.

Countries, product profile, patent watch and counsel notes: the rest of the FTO screens

In shortEverything you can open, in the order a team uses it. Each number traces to a claim, a document or a register entry. Settings hold the weights and say who may sign.

Five AI agents search and read; patent counsel gives every conclusion and signs

In shortEach agent has one job and lines it never crosses. None may write a conclusion, a recommendation or a "freedom to operate" statement. Named patent counsel signs. Your outside patent attorney can use the pack to give a formal opinion.

Search builder, claim reader, element matcher, status checker, landscape summarizer: their limits

CMC owns the facts, patent counsel signs, and your attorney gives the opinion

What the agents may not do: judge validity, contact patent owners, file anything with an office, or change the product profile. Every view and export goes on the audit trail, access is limited by role to the patents team and named CMC leads, and business development sees signed conclusions only. How privilege applies to your use is your counsel's call, and US law on AI tools and privilege is unsettled.26,52

The head of patents, analyst, CMC lead and deal lead in the MRW-217 sample

From a product profile and patent records to a ranked claim list, cited charts and a signed pack

In shortThe solution reads what you already have: a product's documents, patent full text, registers and court records. It returns a ranked list, claim charts, a country-by-expiry map, design-around scenarios and a pack for counsel. The outcome numbers below are design targets, to be measured in your own use.

Connecting patent data feeds, document systems and docketing

Days per update and claims left to read: the design target and the sample run

Those targets say nothing about the quality of a conclusion. Counsel's two days are the part that stays. Security: SOC 2 Type II audited, and information security managed to ISO 27001.

What a blocking claim found late costs you: opinion fees, a patent suit and, above all, delay

In shortA comprehensive opinion costs tens of thousands of dollars, one vendor says. A patent owner's small suit through appeal costs about $1 million, and each day of delay costs far more than the search that could have found the claim. The case for changing the work rests on when you find the claim. Saved analyst days are the smaller part.

Replaying your last FTO update to find your own numbers

Partner diligence, lost options and a thin record

Your return on FTO automation: analyst days, earlier catches and counsel's time on what matters

In shortSix things drive the value. Analyst time is the easy one to count. An earlier catch is the big one, and it is rare, so the calculator treats it as a chance per year. Put in your own numbers. If you run few programs, expect time alone not to pay.

Your days back on six programs, and what one earlier catch is worth

A biotech with one antibody and no near lock date should expect the days not to pay, and should decide on the earlier catch alone. Whether that is worth it depends on your lock date and what CMC says a late change costs.

Is your IP team ready to automate FTO reading? Score your programs, data and sign-off

In shortPick the answer closest to your team today. Eight questions on your programs, product profile, patent data, sign-off and privilege score 0 to 2; the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

FTO software vendor questions: how it reads a claim, whether it refuses an opinion, how it keeps privilege

In shortAsk us the same questions. A good answer shows you, on your own patents, with a citation you can open. Ask to see a "does not read", because that is where a mistake costs most.

We found no published independent test of any FTO tool, as of November 2026. A replay on your own closed update is the most honest evidence you can get, from us or anyone else.

A ten-week FTO pilot: replay a closed update, score it against counsel, then run one live program

In shortStart with an update counsel has already reviewed, so you have an answer key. Measure what the tool finds and misses. Then run one live program with counsel signing, and decide with numbers in hand.

Claims missed, false reads and days per update: what to record in the pilot

Freedom to operate: answers to what IP teams and biotech founders search for

In shortEach answer stands on its own. Legal status is as of November 2026, and none of this is legal advice.