“Physical AI” is the phrase venture investors have been rehearsing all year — machines that perceive and act in the real world rather than just generating text. Most of it is still robots on stage and demo reels. Hakimo, a Menlo Park startup, is one of the few putting the label to work in a live, revenue-generating deployment: pointing AI at the security cameras a building already owns and asking it to do the tedious, error-prone job of watching them.
On July 8, the company said it had closed a fresh round to scale that bet. Below is what the money buys, why security monitoring is turning out to be an unusually clean first market for physical AI, and the harder questions — about surveillance, error and consent — that any honest look at this category has to sit with.
The raise
Hakimo raised $12 million in a growth round led by Zigg Capital, with continued participation from Vertex Ventures, Neotribe Ventures, Rocketship.vc and Defy.vc, according to the company’s own announcement and coverage in Entrackr. The deal takes Hakimo’s total funding to roughly $32 million, building on the $10.5 million Series A it closed in March 2025 to launch its flagship product.
That product is what Hakimo calls the AI Operator — an autonomous agent that ingests live feeds from a site’s existing cameras, alarms and door sensors, and is trained to flag anomalies such as tailgating, loitering, intrusion into restricted zones or unauthorised access. The pitch is deliberately un-flashy: no rip-and-replace hardware, no new camera network, just a software layer on top of infrastructure buildings already paid for. The AI triages what it sees, escalates the events that look like genuine threats, and hands a human operator a much shorter, higher-signal queue.
Hakimo reports that over the past year it tripled revenue year over year and grew to more than 300 customers spanning Fortune 500 enterprises, commercial real estate and multifamily housing. It claims customers see up to a 60% reduction in security incidents and that a single operator can now cover what previously took ten. Those figures come from the company and have not been independently audited, so treat them as vendor-reported rather than benchmarked — but the direction of travel is consistent with what the round is funding.
Why physical AI is deploying here first
It is worth asking why security monitoring, of all the things physical AI could do, is where a deployable business is emerging first. The answer is that the work has three properties that suit today’s models almost perfectly.
It is high-volume and repetitive. A monitoring centre may watch hundreds or thousands of camera feeds where, for hours on end, nothing happens. Human attention degrades fast under exactly those conditions — the well-documented problem of a guard missing the one event that mattered on screen 47 at 3 a.m. Software does not get bored, and narrow visual anomaly detection is a task current computer vision genuinely does well.
It has clear, countable ROI. False alarms are the tax on the whole industry: they burn guard hours, desensitise operators and, at scale, draw fines from police departments tired of responding to nothing. A system that suppresses noise and compresses ten operators’ work into one has a payback a facilities manager can put in a spreadsheet. That is rare in AI, where value is often diffuse and hard to attribute.
And it offers a concrete commercial wedge. Because Hakimo sits on top of existing cameras, the cost and friction of adoption are low, and the buyer — a security or real-estate operator — already has a budget line for exactly this. Physical AI does not need to invent a market here; it needs to do an existing, unglamorous job more cheaply. That is usually how a general-purpose technology actually lands: not with the moonshot, but with the boring workflow nobody wanted to staff.
The uncomfortable questions (handled squarely)
None of that dissolves the obvious tension. A cheaper, tireless, always-on watcher for camera networks is also, described from a different angle, cheaper and more pervasive surveillance. It is worth being precise about the risks rather than waving them away.
Surveillance creep and normalisation. The economics that make Hakimo attractive — low cost, no new hardware, one operator doing the work of ten — are exactly the economics that let monitoring expand into places that could never previously justify it. When watching everything becomes cheap, the default drifts from “monitor the loading dock” toward “monitor everywhere, continuously.” The technology is neutral; the incentive gradient is not.
Bias, error and the cost of false positives. Reducing false alarms in aggregate is good. But a false positive in security is not a spam email — it can mean a real person flagged as a threat, approached, questioned or worse. If detection accuracy varies across skin tone, clothing, disability or the simple fact of loitering while poor, the harm lands unevenly on people who did nothing wrong. A 60% cut in incidents tells you nothing about who the remaining errors fall on. That distribution, not just the headline rate, is what deployments owe the public.
Consent, data governance and oversight. AI monitoring raises questions a passive DVR never did: who can query the footage, how long behavioural inferences are retained, whether people in a monitored space know an algorithm is judging their movements, and who is accountable when it is wrong. These are governance choices, not technical ones — and the honest position is that the industry, Hakimo included, is being built faster than the norms around it. To its credit, the company frames the AI as escalating to human operators rather than acting autonomously on people; keeping a human in that loop is the right instinct and worth holding vendors to.
The India read
The India angle here is real but should be stated carefully. Hakimo is a US company, headquartered in Menlo Park — not an Indian startup. What connects it to India is its founders: CTO Sagar Honnungar grew up in Bengaluru and studied at IIT Madras before Stanford, and the founding team draws on the familiar IIT-to-Silicon-Valley pipeline. That is a diaspora story, not a domestic-market one, and it is worth not overstating it into an “Indian company” headline it isn’t.
The more useful India read is about the technology, not the cap table. India is one of the most camera-dense environments in the world — from residential societies and metro networks to city-wide programmes — and the same physical-AI monitoring that reduces false alarms in a US office tower will be commercially tempting here, at far larger scale. The upside is genuine: better response in genuinely under-staffed security contexts, from campuses to transit.
But India arrives at this moment with weaker guardrails than the deployment deserves. The Digital Personal Data Protection Act is still bedding in, public-space video surveillance operates with limited independent oversight, and the norms around algorithmic monitoring of citizens are thin. The risk is not the technology itself but importing the capability ahead of the governance — consent standards, retention limits, audit rights, and clear accountability when the system is wrong. The safety benefits are worth having. They are worth having with the civil-liberties protections wired in from the start, not retrofitted after the cameras are already watching.
Reporting in this piece — the funding, investors and company-stated metrics — is sourced and attributed above. The analysis of why the market is forming here, and the sections on surveillance and India’s guardrails, are the view of zoho.social’s editorial desk.
