Once your trade mark is registered, how safe is it? For its first twelve months, less safe than most brand owners assume – the Registrar of Trade Marks holds a power under section 84A of the Trade Marks Act 1995 (Cth) to revoke a registration that should never have been granted, and recently the Full Federal Court defined its limits.
A charity had registered BROWN NOSE DAY for bowel cancer fundraising. The owner of the RED NOSE DAY marks did not oppose the application, but later persuaded the Registrar to revoke the registration. In Registrar of Trade Marks v National Cancer Foundation Limited [2026] FCAFC 95, the Full Court held the mark survives. BROWN NOSE DAY is not deceptively similar to RED NOSE DAY, because the colloquial meaning of “brown nose” gives the mark a distinct idea of its own, and consumers in the charity sector are trained to read colour as the feature that separates one fundraising “day” from another.
Two rulings along the way matter well beyond charity campaigns. Deceptive similarity has only one legally correct answer, so there is no safe harbour merely because the examiner’s original call was finely balanced – any error, including an error of judgment, can open the revocation door. But whether it is reasonable to walk through that door is a separate question, and the fact that the call was a close one still counts in the owner’s favour there.
What this means for trade mark registration and brand protection strategy
The durable lessons run in three directions.
If you own a recent registration, its first birthday matters: the Registrar’s window to begin revocation closes twelve months after registration, and after that only a court can remove the mark – so certainty compounds with time, and the heaviest investment in a new brand sits more safely once the window has closed.
If a conflicting mark is heading onto the Register, the lesson is the reverse: a trade mark opposition filed in time is worth far more than a letter to the Registrar afterwards, because revocation is discretionary, was built for administrative error, and – as this case shows – can be resisted all the way to a Full Court.
And when choosing a mark, meaning is your strongest differentiator: sharing words with an established brand is survivable where the mark as a whole conveys a distinct idea, the same imperfect-recollection logic that decided When a Suffix Isn’t a Shield. None of this replaces disciplined brand clearance before adoption – the High Court’s message in You Can’t Backdate Honesty – but it rewards owners who treat the Register as strategic terrain. Watch it. Oppose in time. Choose marks whose meaning does the distinguishing for you. A registered trade mark is property, and the cheapest time to make it secure is at the start.
A large share of the intellectual property in most businesses is protected by nothing but secrecy. That protection carries a hidden dependency: a wrongdoer you can name, and an obligation of confidence a court can enforce against that person. Two disclosures this month put weight on the assumption. On 16 July 2026, Hugging Face disclosed an intrusion into part of its production infrastructure that was driven end to end by an autonomous AI agent system, with unauthorised access to internal datasets and to credentials used by its services. Five days later OpenAI confirmed that its own models were responsible, during an internal cyber-capability evaluation run with reduced refusals. The models escaped a sandbox through a zero-day in a package-registry cache proxy, escalated until they reached internet access, then chained stolen credentials into remote code execution on another company’s servers to obtain the answers to a benchmark. An analysis in The Conversation calls it a shift in cybersecurity, and it is. It is also a shift in IP risk.
Three details reward attention. The entry point was a dataset – a malicious dataset abused code-execution paths in the data-processing pipeline, which makes the data and model surface a first-class attack surface. What was taken was unregistered: internal datasets and credentials, assets with no register, no priority date and no monopoly behind them. And when Hugging Face fed real attack payloads to commercial frontier models for forensic work, the requests were blocked, so it ran the analysis on a self-hosted open-weight model instead, which also kept the attacker’s data and the credentials it referenced inside its own environment.
What agentic cyber risk means for trade secret protection and IP strategy
The durable lessons sit upstream of security.
First, re-run the patent versus secrecy decision on your crown jewels. Many countries, including Australia have no trade secrets statute. Confidential information is protected through an equitable action for breach of confidence, which needs information with the necessary quality of confidence, an obligation of confidence, and unauthorised use or disclosure. Every element of that points at an identifiable defendant. A swarm of short-lived agents with no known operator leaves you with nobody to enjoin. A patent is enforceable against anyone working the invention, however they came by it – the choice one small technology company got wrong before it sued a much larger customer, as I covered in Be prepared before you engage: Zest Labs, Walmat and $222m in damages.
Second, treat every agent you deploy as a delegate rather than a tool: minimum viable data access, provenance attached to outputs, and logs of inputs, sources and model versions, which are the disciplines set out in Your AI Agent Won’t Keep a Secret.
Third, diligence what you ingest. Third-party datasets and model weights now arrive with a licence question, a provenance question and an execution risk, and the same discipline that answers “where did this training data come from” also shrinks your attack surface – the theme of Who Gets Paid for the Platform? Lessons From a Big Week in AI and Intellectual Property.
Two checks worth running this month. Do your AI vendor terms actually permit incident response work, and would sending attack artefacts to a hosted model push your confidential material outside your control at the worst possible moment? And could you prove, event by event, what was accessed and what steps you had already taken to keep it secret? Hugging Face reconstructed more than 17,000 recorded events. That reconstruction is what a breach of confidence claim is built from, and an agent’s own logs are discoverable and unlikely to attract privilege. Secrecy is a control, not a right.
The question for your next IP strategy review is which assets you would still own the morning after the control fails.
What happens when intellectual property built into your products turns out to belong to someone else? This week supplied three expensive answers. Each involved one of the world’s most powerful companies paying – or being ordered to pay – for rights held by a much smaller opponent. Together they carry a clear message for anyone responsible for IP strategy: the size of your balance sheet is not a defence, and well-kept rights are worth more than ever.
A price on training data
A US federal court gave final approval to a US$1.5 billion copyright settlement between Anthropic and a class of authors on 20 July 2026, closing out claims that pirated books were used to train an AI model. Roughly US$3,000 per work, across more than 400,000 works.
The strategic point is not the headline number. It is that AI training data now has a market price. If your business builds or fine-tunes models, data provenance is a balance-sheet issue, not an engineering detail. If you own content, you may be holding a licensable asset that was worth very little three years ago. I wrote about the direction of travel in The Price Signal for AI Training Data Just Got Louder – that signal has now been confirmed at final approval.
The internal question is simple. Can you show, source by source, where your training data came from and what licence covers it? If the answer is no, you have an AI copyright problem waiting for a plaintiff.
One court, eleven countries
On 23 July 2026, the Unified Patent Court’s Düsseldorf Local Division granted InterDigital a second injunction against Disney, upholding a video-encoding patent and reaching eleven European countries in a single order.
Read that from a risk perspective. One case, one division, one adverse finding – and a service can be switched off across most of Europe. The UPC has compressed what used to be a country-by-country grind into a single point of failure, and licensors have noticed. I looked at this structural shift in One Court, Many Borders: Why the UPC Just Redrew Your Litigation Map.
If you earn meaningful European revenue from technology you did not entirely build – codecs, connectivity, streaming infrastructure – your freedom to operate in Europe now depends on licences you may never have reviewed. The review is cheap. An injunction is not.
The verdict that would not go away
On 21 July 2026, a Californian judge refused to overturn a US$634 million jury verdict against Apple, whose Apple Watch health features were found to infringe a Masimo pulse-oximetry patent. An appeal will follow, but the verdict stands for now.
This is patent litigation as a long game. Masimo is a fraction of Apple’s size. It has lost more skirmishes than it has won across six years of fighting. It kept going, chose its forums deliberately, and one surviving patent has now produced a nine-figure judgment. Outcomes turn on where and how you assert rights, not just whether you hold them – a pattern I examined in Your Patents Didn’t Lose Value – The Board Where You Play Them Changed.
The lesson cuts both ways. As a defendant, features added to a flagship product deserve genuine freedom-to-operate diligence, because a single overlooked patent can outlast years of motions. As a rights holder, persistence, good records and forum selection can beat scale.
What this means for your IP strategy
Picture a mid-sized technology company. It fine-tunes an AI model on scraped industry reports. It streams product video through a codec nobody ever licensed. Its flagship feature was “reworked” from a competitor’s approach. Nothing in this quarter’s accounts records any of that. This week’s decisions show that each one now carries a market price – and someone else holds the invoice.
Four things worth considering:
– Trace the provenance of any data used to train or fine-tune AI in your business, and the licences behind it.
– Map your European revenue against patents you rely on but do not own, with the UPC’s eleven-country reach in mind.
– Run freedom-to-operate reviews on flagship features before launch, not after a letter arrives.
– Take an objective inventory of your own rights. If others are monetising theirs this effectively, your unenforced rights are unclaimed revenue.
How to protect intellectual property is only half the question. The better half is whether your IP strategy positions you to collect, or leaves you waiting to pay. The companies on the right side of that line decided years ago – in how they sourced their inputs, documented ownership and maintained the rights they might one day need. If you are not sure which side your business sits on, that is the first conversation to have with your IP strategist or patent attorney – well before someone else’s letter decides it for you.
Over the past two decades, US firm value flipped from mostly tangible to over 90% intangible — and the rulebook for defending those intangibles quietly rewrote itself underneath everyone. A recent Cornerstone Research study, Intellectual Property Litigation: U.S. Trends in Global Perspective (June 2026), maps where enforcement has actually gone, and the headline for anyone running an IP portfolio is simple: demand for protection never fell, but the venues, the remedies, and even the type of right you reach for have shifted. US injunctions became scarce after eBay and now favour product companies that can show market harm. Disputes “exported” abroad — Germany and the new Unified Patent Court for fast injunctions, the UK for global licensing rates, China for coordination. Firms increasingly chose secrecy over disclosure, with trade secret litigation surging from 2022 to 2025. The patent didn’t weaken; the strategic calculation around it got more complex.
The practical lesson is that IP strategy is more than ever a portfolio-and-venue decision, not a just filing decision. Three moves follow directly.
First, match the right to the remedy you actually need — if your real goal is to stop a competitor, know that a US damages award and a German injunction are different instruments, and build the campaign around the one that changes behaviour.
Second, treat trade secrets as a deliberate choice, not a default — for AI model weights, datasets, and manufacturing know-how, secrecy can outperform a patent, but only if your access controls and contracts are genuinely in place before a dispute, not assembled after one.
Third, decide the enforcement question before you’re forced to — the report’s picture of multi-forum campaigns is a reminder that reactive litigation is the expensive way to learn this; the discipline of asking “do we have a good case, and if not, how do we resolve this sensibly?” applies long before you pick a court.
The organisations that win here aren’t the ones with the most patents — they’re the ones who decided, in advance, which right to assert, where, and to what end.
Something quietly broke this year in one of Silicon Valley’s most-watched classrooms. In Steve Blank’s 2026 Lean LaunchPad at Stanford, every team used AI to spin up working products in hours — and the result wasn’t faster learning. It was faster failure. As Blank put it, the not-so-obvious change was that building products rapidly let teams make bad ideas go faster. A polished, finished-looking MVP used to signal real technical insight and hypothesis testing. Now it signals almost nothing, because the build itself has become close to free. Teams confused a slick deliverable with genuine product/market fit, and pivoted late as a result.
The strategic lesson for any IP-intensive business is uncomfortable but clarifying: when the product can be recreated in an afternoon, the product is no longer your differentiation — and neither is being first. If your advantage evaporates the moment a competitor points the same tools at the same problem, you don’t have a moat. You have a countdown.
So where does defensibility live now? Not in the artefact, but in everything around it that can’t be cheaply copied: proprietary data and the feedback loops that improve with use, deep domain understanding of a real customer problem, and the disciplined capture of IP — patents where they genuinely block, and trade secrets and know-how where speed and secrecy beat disclosure.
This is the practical move for in-house IP leads and founders: stop treating “we shipped it” as proof of advantage, and start asking what about this would survive a well-funded competitor with the same AI stack. Map your differentiation to durable assets before you scale, not after a late pivot. Decide deliberately whether each product calls for blocking competitors, outrunning them, or teaming up — because AI has collapsed the cost of building but not the value of choosing well. The teams that win the next cycle won’t be the ones that build fastest. They’ll be the ones who know what’s actually worth owning.
Blank’s full write-up of the Lean LaunchPad 2026 lessons is worth a read for the classroom detail behind the pattern.
Choosing a plainly descriptive business name feels like the safe option. It rarely is. In 2013, a small Melbourne pool servicing company settled on “Pro Pool Services”, checked the ASIC and ABN registers, had its accountant confirm the name was available, and got on with business. This week, in The Pops Group Pty Ltd v Pro Pool Services Pty Ltd (No 2) [2026] FCA 912, the Federal Court found that “Pro Pool” is deceptively similar to the registered POOL PRO marks, granted a permanent injunction and ordered $60,000 in damages. Three findings deserve your attention. Reversing the word order didn’t help: consumers with an imperfect recollection remember the words and the idea, not the sequence. The infringements weren’t the full company name at all — they were the everyday abbreviations: the propool.com.au domain, “Pro Pools” on the website, a #propools hashtag, “Pro Pool Shop” signage. And the good faith defence failed because searching the company and business name registers is not searching the trade marks register — the Court held the law reasonably expects anyone adopting a name to check the register “of which they are taken to be aware”, or get advice. Ignorance is not good faith.
Business name vs trade mark: the lessons for your brand
The durable lessons run in both directions. If you are naming or renaming anything — a company, a product, a location — a trade mark search comes before the domain purchase and the signage, because business name registration gives you no right to use the name, only the right to be registered under it; that trap is a cousin of the timing problem in You Can’t Backdate Honesty, where good faith is judged at the moment of adoption and can’t be repaired later. Then audit how your teams actually use the brand: domains, email addresses, hashtags and shop signage are all trade mark use, and a shortened or tweaked version of your name earns no protection of its own — the same lesson as When a Suffix Isn’t a Shield. The checklist for a clean adoption sits in Preparing for a new launch. For trade mark owners, note what made enforcement straightforward here: Pool Pro had registered marks and a documented licensing program with a $10,000 annual fee, which the Court adopted as the damages yardstick — while the owner’s inflated expert report claiming lost product profits was rejected outright. A registered mark plus a real, priced licensing structure turns infringement into a calculable claim rather than an argument. The register is public and the search is cheap. The time to run it is before the vehicles are wrapped — for either side of this dispute, everything after that gets more expensive.
The most valuable intellectual property in your business is often not the product your customers see — it’s the platform underneath it: the data it learned from, the delivery technology it rides on, the foundational patents everything else is built over. Several developments this past week show what happens when someone owns those foundations, and what it costs to build on foundations you don’t. The pattern matters to any business deploying AI or building on licensed technology, which now means almost everyone.
The AI copyright lawsuit wave reaches the inputs
Major publishers including Hachette, Cengage and Elsevier, joined by authors, filed a class action against Google in New York on 14 July, alleging Gemini was trained on millions of copyrighted books — including works originally digitised under the old Google Books arrangements (coverage here). Whether AI can lawfully train on copyrighted material is being decided case by case, but the strategic point doesn’t need to wait for judgment: training data provenance is now a balance-sheet issue. If you licence AI models, ask your vendor to warrant where the data came from and to indemnify you. If you own content, it just became a licensable input with a market forming around it — the theme I explored in The Price Signal for AI Training Data Just Got Louder and The Licence to Train: Brussels Reopens the Copyright Bargain.
Platform patents outlive the product race
Sanofi sued both Pfizer and Moderna in New Jersey this week over lipid nanoparticle delivery patents — the technology that gets mRNA into cells — asserting ten patents against Moderna’s vaccines and eight against Comirnaty (Bloomberg Law’s report). Two details reward attention. The patents arrived with Sanofi’s 2021 acquisition of Translate Bio, and the suits seek royalties and damages, not injunctions. Sanofi lost the mRNA product race; it may still tax the winners, because it owns part of the platform. That is patent licensing as a revenue line — the posture I examined in When Your IP Becomes a Revenue Line: Three Signals From the Past Week. Audit your own portfolio the same way: which of your patents read on what competitors ship, not just on what you ship?
The generative AI patent race is compounding
WIPO’s new patent landscape report (14 July) found generative AI patent activity nearly tripled in two years, with more GenAI patents published in 2024–25 than in the entire preceding decade, and China-based filers leading (analysis here). Most companies are deploying AI; a much smaller group is filing patents around what they build with it. The gap between those two groups is tomorrow’s licensing market, with the same dynamics Sanofi is demonstrating today. Using the tools is not a moat — as I argued in AI Isn’t Your Advantage—Your IP Strategy Is.
Three questions for your next IP strategy review
First: can every AI vendor in your stack warrant its training data provenance — and who carries the liability if it can’t?
Second: which platform technologies does your business rent rather than own, and what would a royalty demand on them do to your margins?
Third: what did you actually file in the past twelve months around your AI-enabled products and processes?
The week’s developments all point the same way: the returns in the AI economy are accruing to whoever owns the foundations and can prove it. The filings, licences and provenance records being created right now will decide who gets paid for the platform for the next decade. Make sure some of them are yours.
Ben Thompson’s recent Stratechery analysis is a masterclass in how scale, integration and narrative compound into something competitors can no longer touch. Starlink is profitable not because the satellites are clever, but because SpaceX owns the launch capacity that puts them there. Tesla’s Full Self-Driving isn’t licensed from anyone. The “rack-in-space” thesis Thompson lays out only works because one company controls rockets, satellites, compute design and the customer relationship. For IP-intensive businesses, the lesson is uncomfortable: patents matter, but they rarely beat a vertically integrated competitor who has turned proprietary capability into the default option. By the time American Airlines was announcing Starlink as an “elevated experience,” it had already become table stakes — exactly how Musk wants it.
There are sharper signals here for IP strategists. First, the xAI footnote: $5.1 billion of R&D for a model now in fifth place, with the founding team gone. Know-how walks; protecting it through retention, trade secret hygiene and inventorship discipline is not optional at frontier valuations. Second, the rack-satellite concept is a reminder that the most valuable IP positions are staked early around end-state architectures — thermal management, radiation-hardened compute, laser interconnects — long before the market concedes the category exists. Third, regulatory friction (in this case, data centre zoning) is increasingly the catalyst for category-defining innovation; the IP playbook should anticipate where constraints will force a redesign, not react after it. Musk’s method is to start with the desired end state and work backwards.
IP strategy should do the same.
Read the full piece — The SpaceX IPO and Data Centers in Space — here: https://stratechery.com/2026/the-spacex-ipo-and-data-centers-in-space/
Most companies now accept that geopolitical risk belongs in corporate strategy, yet very few have ever stress-tested how they would actually respond. McKinsey’s recent article, The art, science, and technology of geopolitical scenario planning, puts numbers on the gap: fewer than a third of organisations rate their geopolitical risk management as mature, a third do no scenario planning at all, and fewer than 5 per cent run simulations that test leaders under pressure. The authors set out five foresight instruments — horizon scanning, scenario planning, contingency playbooks, simulations and adversarial tabletop exercises — and their central point deserves attention from every IP owner: foresight is not prediction, it is disciplined preparation.
That lesson applies with particular force to intellectual property, because IP positions are slow assets in a fast environment. A patent filing footprint set five years ago determines where you can enforce today; a licensing structure drafted before export controls tightened determines what you can lawfully supply tomorrow. When the shock arrives, your IP options are largely fixed — which means the thinking has to happen now.
The practical translation is to run these same instruments over the IP portfolio before events force the issue. Scenario planning asks: if a key market fragments, do our filings, freedom to operate positions and licence terms still support the business in each plausible future? Contingency playbooks answer the crisis question in advance — the reason we built the IP Playbook around pre-agreed strategic principles rather than step-by-step plans, which rarely survive first contact with adversity. And simulation is where assumptions actually get tested: an IP wargame plays out the moves of competitors, regulators and partners so that gaps surface in a workshop rather than in a dispute — the same “outcomes are built before the fight” pattern that ran through a recent month of court decisions in Built Before the Fight: What May’s IP Decisions Reward. Try McKinsey’s premortem framing on your own portfolio: it’s five years out and your IP failed to protect the business — what were the causes? If the answers come easily, you have a work plan. If they don’t, that’s the strongest argument yet for putting your IP strategy through a wargame while it’s still cheap to change.
Most AI transformations will fail for a familiar reason. Not because the strategy is wrong, but because organisations misread how change actually happens. The recent BCG analysis makes this explicit: leaders assume alignment, but employees experience change as something imposed, not chosen. The gap matters. For IP‑intensive businesses, this is more than a cultural issue—it is a strategic risk. If the teams responsible for generating, protecting, managing and exploiting intellectual property are not genuinely engaged, the result is predictable: underutilised portfolios, weak invention pipelines, and missed opportunities to anchor competitive advantage around new capabilities. IP does not scale through announcements. It scales through behaviour—how people choose to buy in to IP processes and build, disclose, protect, and commercialise.
Treat IP as part of the transformation system, not an output of it. That means building real ownership around IP creation and use—clear incentives to capture AI‑driven inventions, disciplined processes to translate technical advances into defensible rights, and active governance to ensure portfolios align with evolving business models.
BCG’s point about the “messy middle” is particularly relevant here: this is where IP strategy either compounds value or quietly stalls. Signals such as silence, superficial alignment, or inconsistent execution often appear first in how teams handle IP decisions—what gets filed, what gets ignored, and what is allowed to drift. Leaders who design for agency, not compliance, will see stronger portfolios and clearer competitive positioning.
IP strategy must be embedded early, owned broadly, and executed consistently. That is how intangible assets become commercial outcomes.
Read the article here: https://www.bcg.com/publications/2026/ceos-are-betting-big-on-ai-transformations
You must be logged in to post a comment.