AI Physical Security Vendor Due Diligence: The 2026 Market Analysis of Financial-Health, Claims-Substantiation, and Platform-Risk Signals Buyers Must Verify Before a Multi-Year Contract

Why AI physical security vendor due diligence now turns on financial signals, not just detection accuracy

AI physical security vendor due diligence has become a financial exercise as much as a technical one, because the market has reached the stage where the product usually works and the company selling it sometimes does not. A security director evaluating an AI video analytics platform in 2026 can run a proof of concept, watch the system flag a brandished weapon with a confidence score on screen, and still sign a five-year agreement with a vendor that will not survive to year three. Detection performance has become table stakes. The harder question, and the one most procurement processes never formally ask, is whether the corporate entity behind the model has the financial runway, the substantiated claims, and the verifiable third-party signals to be a durable platform partner rather than a counterparty risk.

This market analysis treats vendor selection as an underwriting problem. It lays out a four-layer due-diligence stack that risk officers, procurement teams, and security directors can apply to any AI physical security vendor before committing budget: financial health, claims substantiation, platform and consolidation risk, and the verifiable trust signals that separate a marketing deck from a defensible record. The framing is deliberately vendor-neutral. The same questions that protect a buyer from a startup with twelve months of cash also protect them from an established integrator quietly being absorbed into a roll-up. The goal is to give the non-financial security buyer a structured way to read the company, not only the camera feed.

The market's headline numbers cannot anchor a vendor decision

The first signal that the AI video surveillance category is still loosely defined is how violently its own size estimates disagree. Across published 2026 forecasts, the market is valued anywhere from roughly four billion dollars for narrowly scoped "AI in video surveillance" to nearly twenty-eight billion dollars for broad "video analytics," a spread of about 6.8 times depending entirely on where each analyst draws the boundary. MarketsandMarkets, for example, projects the AI-in-video-surveillance segment growing at a 17.9 percent compound annual rate toward roughly 10.9 billion dollars by 2032, while broader video analytics definitions land at multiples of that figure for the same period. A category whose total addressable market can be stated as a 6.8x range is a category where definitions, not measurements, are doing the work.

For a buyer, the practical takeaway is not which number is right. It is that the headline figures a vendor cites in a pitch deck are selected, not discovered, and that the same definitional looseness that inflates market-size slides also inflates accuracy claims, deployment counts, and competitive comparisons. Market enthusiasm is real and the growth is genuine, but enthusiasm funds entrants faster than it filters them. The result is a crowded field in which capital availability, not demonstrated durability, determines who is still standing at contract renewal. That is precisely the environment in which disciplined due diligence stops being a procurement formality and becomes a loss-prevention control.

None of this argues against adopting AI physical security. The operational case for moving from passive recording to real-time detection is well established, and IntelliSee has documented the economics of that shift in its four-variable ROI framework. The argument here is narrower and complementary: once a buyer has decided that AI detection is worth funding, the next analytical task is to evaluate the vendor as a going concern, not merely as a demo.

Layer one: financial health and the going-concern signal

Financial durability is the layer security buyers are least trained to assess and the one that most often decides whether a contract delivers its full value. The structural reality is unforgiving: roughly 38 percent of venture-backed startups exhaust their cash before reaching eighteen months of operation, and a 2025 review of the funding environment found that 61 percent of startups saw their runway shrink year over year as follow-on capital became harder to raise. The Series A conversion rate, the share of seed-funded companies that secure their next institutional round within two years, has fallen to roughly 15 percent from more than 30 percent in the prior cycle. A buyer signing a multi-year agreement with an early-stage AI security vendor is, whether they frame it this way or not, making a bet on that company's next fundraise.

The most useful single artifact for assessing this is the going-concern disclosure. Under the U.S. accounting standard FASB ASC 205-40, management is required to evaluate, every annual and interim reporting period, whether there is substantial doubt about the entity's ability to continue as a going concern within one year of the financial statements being issued, and to disclose the conditions, its evaluation, and its mitigation plans when that doubt exists. For any publicly traded vendor, those disclosures are filed and freely searchable through the U.S. Securities and Exchange Commission's EDGAR system. A buyer does not need an accounting background to find the phrase "substantial doubt" in a 10-K, and its presence is one of the clearest viability flags available without an NDA.

Private vendors, which make up most of the AI physical security field, do not file with the SEC, so the buyer's leverage shifts to the procurement table. Reasonable, defensible diligence requests for a private vendor entering a material multi-year contract include audited or reviewed financial statements under NDA, the most recent funding round and its date, the named lead investors, and a direct question about months of runway at current burn. A vendor confident in its durability treats these as ordinary enterprise-procurement questions. A vendor that deflects them is answering the question by declining to. Two adjacent signals deserve equal weight: revenue concentration, where a vendor dependent on one or two marquee accounts carries hidden fragility, and the ratio of research-and-development spend to revenue, which indicates whether the roadmap being sold is actually funded.

This is also where the economics connect to the rest of the buying calculus. The total cost of ownership of an AI physical security deployment is incurred over the full lifecycle, not at signing, which means a vendor's ability to fund support, model retraining, and integration maintenance over that lifecycle is a direct input to whether the buyer ever realizes the return. A cheaper license from a vendor with twelve months of runway is not cheaper. It is a deferred cost with a probability attached.

Layer two: claims substantiation and the FTC reasonable-basis standard

Every objective performance claim a vendor makes is, in principle, a representation it must be able to back with evidence. The Federal Trade Commission's longstanding advertising substantiation doctrine holds that it is an unfair and deceptive act under Section 5 of the FTC Act to make an objective product claim without a reasonable basis, consisting of competent and reliable evidence, possessed at the time the claim is made. For health and safety claims the standard rises to competent and reliable scientific evidence evaluated by qualified persons. Detection rates, false-alarm percentages, and response-time figures are objective claims that touch directly on safety, which places them squarely inside the heightened standard. The FTC has reinforced this expectation in recent years by issuing Notices of Penalty Offenses to hundreds of companies, a clear signal that unsubstantiated performance marketing is an enforcement priority, not a gray area.

For the buyer, the regulatory standard converts neatly into a procurement test. When a vendor cites a detection accuracy figure, the right follow-up is not "how did you measure that" asked casually but a documented request for the underlying evidence: the test methodology, the dataset, the conditions, who conducted it, and whether the result is reproducible on the buyer's own camera footage and lighting. A claim that cannot be reproduced on your environment is a claim about the vendor's environment. This is the connective tissue between substantiation and a rigorous proof of concept, a discipline IntelliSee has detailed in its guide to evaluating an AI gun detection system. The POC is where substantiation stops being a slide and becomes a measurement.

What a substantiated detection actually looks like is concrete. It is a bounded object, a class label, and a confidence score produced on a real frame, in real conditions, that a buyer can inspect rather than infer.

The substantiation layer also exposes a quieter risk: claims that are technically true but operationally misleading. A vendor may report a high detection rate measured under ideal lighting and unobstructed sightlines while saying little about performance under occlusion, low light, or adversarial conditions, which is where real deployments live and where computer vision genuinely struggles. A buyer who has read IntelliSee's analysis of how models handle occlusion, low light, and adversarial conditions is equipped to ask the question that separates a marketing number from an operational one: under what conditions was this measured, and what happens to the figure when conditions degrade.

Layer three: platform and consolidation risk

The third layer asks what happens to the buyer's deployment when the vendor's ownership or strategy changes, an outcome that is increasingly likely rather than hypothetical. The AI physical security sector is consolidating, with strategic acquirers and private-equity-backed roll-ups absorbing point-solution vendors at a steady pace, a dynamic IntelliSee mapped in its market analysis of vendor roll-ups and platform risk for buyers. Acquisition is not inherently bad for a customer. It can bring capital and stability. But it routinely changes pricing, support models, integration priorities, and roadmap, and the buyer who signed with the independent company is rarely the priority after the deal closes. Platform risk is the risk that the entity you contracted with is not the entity you end up depending on.

The defense against platform risk is portability, and portability is contractual and architectural rather than aspirational. Architecturally, a vendor that analyzes standard camera streams over open protocols such as ONVIF and RTSP, rather than locking the buyer into proprietary cameras or a closed video management system, preserves the buyer's ability to change course. IntelliSee's technology briefing on retrofit architecture and VMS integration details why an add-on inference layer over existing infrastructure carries materially lower lock-in than a rip-and-replace platform. Contractually, the buyer should secure data-export rights, clarity on who owns configuration and tuning, defined continuity-of-service terms in the event of an acquisition or wind-down, and source-code or model escrow where the deployment is mission-critical. These are ordinary enterprise-software protections that physical security procurement has historically underused.

Consolidation risk also intersects with how the system is bought. Public-sector and large enterprise buyers using cooperative purchasing vehicles, which IntelliSee analyzed in its report on cooperative purchasing in AI physical security, gain procurement speed but should confirm that the awarded entity, and any successor after a merger, remains bound to the contract's terms. The vehicle compresses the buying cycle. It does not, by itself, insulate the buyer from what happens to the vendor afterward.

Layer four: the verifiable trust signals

The final layer is the shortest to check and the hardest to fake, because it relies on validation that an independent authority has already performed. The most consequential of these in physical security is the U.S. Department of Homeland Security SAFETY Act, which provides liability protections to sellers of qualified anti-terrorism technologies and has approved more than 1,000 technologies for coverage since its creation. SAFETY Act status, whether Designation or the higher bar of Certification, signals that DHS has reviewed the technology's efficacy and that the vendor has accepted a level of federal scrutiny most of the field has not. IntelliSee's standards briefing on the DHS SAFETY Act in AI security explains the distinction between the two tiers and why the liability allocation matters to both the vendor and the buyer. For a buyer, the relevant fact is simple: SAFETY Act approval is verifiable on the public SAFETY Act database, so a vendor's claim to it can be confirmed in minutes.

The second verifiable signal is governance maturity, increasingly expressed through alignment with the NIST AI Risk Management Framework. As IntelliSee documented in its NIST AI RMF standards-compliance briefing, procurement teams now expect vendors to produce documentation of how they govern, map, measure, and manage AI risk, and a vendor that can hand over that documentation has demonstrated an operational maturity that a startup running fast and light usually cannot. Add to this the conventional enterprise security attestations, SOC 2 Type II being the baseline, and the buyer has a small set of signals that can be checked against independent records rather than vendor assurances.

The table below consolidates the four layers into a single procurement reference: what to verify, where the evidence lives, and the warning sign that should slow a deal down.

How the four layers compound into a single buying decision

The four layers are not a checklist to be scored independently. They compound. A vendor with strong detection numbers and weak financials is a deferred failure. A financially sound vendor that locks the buyer into proprietary hardware has converted a software decision into a capital decision. A vendor with impeccable governance documentation that cannot reproduce its accuracy claims on the buyer's footage has substituted paperwork for performance. The analytical value of the stack is that it forces the buyer to weigh these together, the way an underwriter weighs the components of a risk rather than approving on any single strong attribute.

Read in sequence, the layers also map to the buyer's actual exposure over time. Financial health governs whether the vendor reaches renewal. Claims substantiation governs whether the system performs as sold on day one. Platform and consolidation risk governs whether the deployment survives a change of ownership. Verifiable trust signals govern whether the buyer can defend the decision to a board, an insurer, or a regulator after the fact. Insurers in particular are formalizing this expectation, and IntelliSee's market intelligence on how carriers underwrite AI physical security shows that the documentation a disciplined diligence process produces is increasingly the same documentation that supports favorable premium treatment. Diligence done well is not only risk reduction. It is an asset the buyer can reuse.

This is where IntelliSee's own positioning fits the framework rather than substituting for it. The platform is built as an add-on inference layer that analyzes a facility's existing camera streams in real time, which keeps the buyer on open infrastructure and lowers platform risk by design. Its detection output is inspectable and reproducible on a buyer's own footage, which is what the substantiation layer requires. Its architecture deliberately avoids facial recognition, continuous video retention, and protected health information, which narrows regulatory exposure. None of that exempts IntelliSee from the same diligence a buyer should apply to any vendor. The point of the framework is that a buyer should be able to run it against every candidate, including this one, and reach a defensible decision on the evidence. To put the stack to work on a live deployment, IntelliSee invites security and risk leaders to request a risk assessment.

Frequently asked questions

What is AI physical security vendor due diligence?

It is the structured evaluation of an AI physical security vendor as a business and a counterparty, not only as a product. Effective due diligence assesses four layers: the vendor's financial health and ability to survive the contract term, the substantiation behind its performance claims, the platform and consolidation risk created by its ownership and architecture, and the verifiable third-party signals such as DHS SAFETY Act status that an independent authority has already validated.

How can a buyer check an AI security vendor's financial stability?

For publicly traded vendors, going-concern disclosures and risk factors are filed and searchable on the SEC's EDGAR system, where the phrase "substantial doubt" under FASB ASC 205-40 is a direct viability flag. For private vendors, which make up most of the field, the buyer should request audited or reviewed financials under NDA, the most recent funding round and date, named lead investors, and current months of runway at present burn. Revenue concentration in one or two accounts and a thin research-and-development budget relative to revenue are additional warning signs.

What does the FTC require for AI detection accuracy claims?

Under the FTC's advertising substantiation doctrine and Section 5 of the FTC Act, a vendor making an objective performance claim must possess a reasonable basis for it, consisting of competent and reliable evidence, at the time the claim is made. Because detection rates and false-alarm figures are safety-related claims, they fall under the heightened standard requiring competent and reliable scientific evidence. In practice this means a buyer is entitled to ask for the methodology, dataset, conditions, and reproducibility behind any number a vendor cites.

What is platform risk in AI physical security?

Platform risk is the risk that the vendor a buyer contracts with is not the vendor it ends up depending on, typically because of acquisition, a roll-up, or a strategic pivot that changes pricing, support, and roadmap. The defense is portability: analyzing standard camera streams over open protocols such as ONVIF and RTSP rather than proprietary hardware, plus contractual data-export rights and continuity-of-service terms that survive a change of ownership.

Why does the DHS SAFETY Act matter when choosing a vendor?

The DHS SAFETY Act provides liability protections to sellers of qualified anti-terrorism technologies, and DHS has approved more than 1,000 technologies for coverage. Because approval reflects federal review of a technology's efficacy and is verifiable on the public SAFETY Act database, it is one of the few vendor claims a buyer can confirm independently in minutes. A vendor holding Designation or Certification has accepted a level of scrutiny that most of the field has not.

Does stronger detection accuracy outweigh a vendor's financial weakness?

No. Detection accuracy determines whether the system performs on day one, but financial weakness determines whether the vendor is present to support, retrain, and integrate the system across the contract lifecycle, which is when most of the total cost of ownership and most of the realized return occur. A high-accuracy platform from a vendor with twelve months of runway is a deferred cost with a probability attached, not a bargain. The layers must be weighed together.

Continue the research

This market analysis pairs with IntelliSee Intelligence reports on the surrounding buying calculus. Read the market analysis of vendor consolidation and platform risk for the acquirer landscape behind layer three, the total cost of ownership report for the lifecycle economics behind layer one, and the proof-of-concept methodology that operationalizes layer two. For the platform itself, see how IntelliSee works and the full solutions overview, or request a risk assessment to apply the diligence stack to your own deployment.