HiddenLayer Patent Portfolio Analysis

The money arrives after the paperwork

A $100M Series B, from Morgan Stanley, Ten Eleven Ventures and Booz Allen Ventures. For a company founded in 2022 that is a short runway to a round of that size. The usual reading is that the capital bought the build. Here the sequence runs the other way. By the time the round closed, the filing record was largely complete and largely granted.

So the pulse on the left is not the start of the story. It is a confirmation of one already written into filing dates at the EPO, the Korean office, the USPTO and via WO. Thirty-four families in roughly three years, from a company whose stated business is protecting AI models and automated workflows from targeted attack. The interesting part is where those claims actually sit.

Twenty-nine granted, five still open

Twenty-nine of the thirty-four families are granted. Five are published and pending. That is an unusual shape for a 2022 company: most portfolios this young are almost entirely applications, with the granted ribbon a thin line. Here the ribbon is thick and the pending band is the narrow one. Granted claims are enforceable claims. They also mean the scope has already been argued with an examiner and survived.

One note on the file. The record shows 29 grants, and separately notes first grants expected around 2027. Those two statements do not sit together, and the resolution matters: it is the difference between a portfolio that has cleared examination and one whose scope is still open to negotiation. [Docket-level confirmation of grant dates per family needed.] Either way, the pending five are the tail, not the body.

Granted claims are enforceable claims. They have already survived an examiner.

Twenty-three families in one year

One family in 2022. Three in 2023. Then twenty-three in 2024, and seven so far in 2025. That single bar in 2024 carries two thirds of everything. It is not a gradual ramp. It reads as a deliberate sweep of a claim space the team had already mapped in the two thin years before it, when adversarial attacks on deployed models were still a research topic rather than a procurement line item.

The shaded window on the right is the part nobody can see yet. Applications publish about eighteen months after filing, so recent filings are absent by definition, not by choice. The 2025 count of seven will rise. Read against the funding pulse, the shape is clear: the heavy filing year came before the round, not after it.

A small circle among very large ones

The claims sit in G06F and H04L, computing and digital communication. Over the last five years those same groups drew Huawei with 2,173 families, Samsung with 1,831, IBM with 1,746 and Microsoft with 1,624, with Tencent Shenzhen and China Mobile Communications close behind. On this scale HiddenLayer's thirty-four is a dot. Count is not the comparison to make, and anyone reading the dot as weakness has read the wrong axis.

G06F and H04L are enormous classifications. Being inside them says almost nothing about overlap. What matters is where inside them the claims land, and the latest filing titles are specific: adversarial attack monitoring for machine learning models, generative model protection using sidecars, multimodal protection using sequential sidecars. That is a narrow shelf in a vast room. Scope, not volume, is the thing to read here.

Dell, Microsoft and IBM point back

Citations run from Dell across sixteen publications, Microsoft across eight, Netskope and IBM across four each, and Kyndryl across four. A young portfolio being cited by companies of that size is the clearest available signal that examiners and drafters treat these families as relevant prior art in the space. It is third-party confirmation of position, arrived at independently of anything the company says about itself.

Now the people. Eighteen inventors are named across thirty-four families, but Andrew Davis appears on fourteen, Kenneth Yeung on fourteen and Tanner Burns on thirteen. Three names carry nearly the whole record. Hiring reinforces the same direction: eighty-one open roles, forty-two in software and AI, including a lead engineer for AI attack simulation. Twenty-seven are go-to-market, which is what a granted portfolio and fresh capital usually buy next.

Three names carry nearly the whole record.

What the record says the money bought

Put the stages back together. A company incorporated in 2022 filed early, swept a narrow claim space in 2024, came out of examination with twenty-nine granted families, and was being cited by Dell, Microsoft, Netskope and IBM before a $100M round arrived. The moat is not the count. It is the timing, the grant rate and the specificity of the sidecar and adversarial-monitoring claims.

The concentration is the open variable. Three inventors on most of thirty-four families is fast execution and a narrow base at the same time. Forty-two of eighty-one open roles are in software and AI, which suggests the base is being widened. There are also five bio and clinical roles, which the filing record does not explain. That corner is worth understanding before assuming the claim space stays where it is.