Humanoid Deployment Risk Allocation: The Contractual Framework Every Industrial Buyer Must Have Before Signing
Updated: 6 days ago
Physical AI Journal Research Team — Independent analysis by Sekason Research Limited, United Kingdom
Report: Humanoid Robot Risk Allocation 2026 | September 2026
Executive Intelligence Synthesis
PHYSICAL AI JOURNAL
Industrial buyers deploying humanoid robots in 2026–2028 face a multi-party risk chain spanning hardware manufacturers, AI software providers, system integrators, and site operators. No standardised contractual risk-allocation model exists for humanoid deployments. Buyers who sign deployment or RaaS agreements without explicitly addressing liability caps, software-update responsibility, supervision obligations, and vendor financial durability are accepting financial exposure that no party has contractually agreed to carry. The EU AI Act’s August 2028 compliance date for high-risk AI in machinery adds a regulatory transition risk to every multi-year contract signed today.
This report answers one decision: whether and under what contractual conditions to deploy a humanoid robot, based on who bears financial exposure when something goes wrong. The five signals below frame the risk environment as of September 2026, when humanoid deployment costs have reached approximately $400,000 per robot over five years (company-claimed, Agility Robotics SEC materials) and commercial contracts are being signed without standardised risk-allocation terms.
Signal 1. The humanoid risk chain is now multi-party by design.
Hardware, AI software, system integration, and site supervision can each fail independently — and no standardised public contract model assigns financial responsibility across all four layers simultaneously. A buyer who signs a deployment agreement without addressing each layer is accepting exposure that no single counterparty has agreed to carry.
Signal 2. 2026–2027 is the contract-design window.
The European Commission has confirmed that high-risk AI systems embedded in physical products — including machinery — carry an application date of August 2028 under the EU AI Act. Buyers signing three- or five-year agreements today are committing to a contract term that spans this compliance transition. Who bears the cost of regulatory adaptation during a live contract is an unresolved procurement question.
Signal 3. The financial stakes are material, not theoretical.
Agility Robotics’ SEC materials present an illustrative five-year ownership cost of approximately $400,000 per robot (company-claimed). At that level, a ten-unit deployment represents a $4 million capital exposure before integration, supervision, insurance, and downtime costs are added. Risk misallocation at this scale carries direct balance-sheet consequences.
Signal 4. Buyers are negotiating in an information vacuum.
No publicly disclosed humanoid deployment contract reveals actual indemnity caps, liability allocation, insurance requirements, SLA terms, or termination rights. The GXO Logistics, BMW, Mercedes-Benz, and Schaeffler deployments — the four strongest examples in the public evidence base — share this absence. Buyers cannot benchmark against prior practice because prior practice is not disclosed.
Signal 5. RaaS shifts ownership — not necessarily risk.
Robots-as-a-Service structures, now a primary commercial model for Agility Robotics with disclosed conditional orders exceeding $300 million (company-claimed), transfer hardware ownership to the vendor. Operational liability, supervision responsibility, and business interruption exposure may remain with the site operator regardless. This distinction is contractual, not structural — and must be confirmed in writing before deployment commences.
The Humanoid Risk Chain: Why Standard Commercial Contracts May Not Protect Industrial Buyers
PHYSICAL AI JOURNAL
A humanoid robot deployment creates a risk chain that can span seven distinct parties: hardware manufacturer, AI or software provider, system integrator, RaaS provider, site operator, employee or contractor, and insurer. A failure can originate from a hardware defect, a perception failure, an AI decision error, a software update, an integration fault, or an operator action. Because each failure mode implicates a different responsible party, a standard equipment purchase contract — which assigns fault to a single vendor — does not adequately protect an industrial buyer from the full exposure profile of a humanoid deployment.
A humanoid deployment routes financial exposure through a seven-party chain that standard equipment contracts were not designed to address — hardware manufacturer, AI software provider, system integrator, RaaS provider, site operator, employee or contractor, and insurer. A conventional equipment purchase assigns liability to one counterparty: the manufacturer built the machine, the buyer operates it, the insurer covers residual exposure, and the chain is short. A humanoid deployment does not work this way.
The hardware and the intelligence that drives it may come from different vendors. System integration — connecting the robot to existing production infrastructure — introduces a third party whose modifications may alter the behaviour the manufacturer warranted. If the deployment is structured as a RaaS arrangement, a fourth party holds hardware ownership. The site operator runs the facility and employs the workers. And a software update issued remotely by the manufacturer after the contract is signed can change the robot’s behaviour without any hardware change occurring.
A Reuters analysis published in May 2026 identified the specific exposure categories this creates: bodily injury, property damage, and business interruption. Munich Re has separately identified physical AI as generating production-shutdown exposures and noted the additional risk of remote hijacking and malware as categories relevant to liability analysis. Neither source found a standardised contractual model in use across the industry.
Figure 1 — The Humanoid Deployment Risk Chain
This multi-party structure creates ten distinct categories of deployment risk that a procurement contract must address. The ten categories are: hardware defect; AI and software failure; software update impact; third-party integration failure; supervision and operator failure; bodily injury; property and equipment damage; business interruption and downtime; cyber incident; and vendor failure or platform discontinuation. Each of these categories can produce a financial loss. In each case, the question that a procurement contract must answer is which party has agreed, in writing, before the robot enters the facility, to bear that loss.
Platform and Market Landscape: Commercial Contracting Has Arrived
PHYSICAL AI JOURNAL
The humanoid robot market moved from technology demonstrations to commercial contracting in 2024–2026. Agility Robotics has disclosed more than $300 million in multi-year Digit v5 conditional orders through both RaaS and ownership models (company-claimed). GXO Logistics signed a verified multi-year RaaS agreement for Digit at its Georgia facility. Mercedes-Benz, BMW, and Schaeffler have entered testing or commercial agreements with Apptronik, Figure AI, and Humanoid respectively. The commercial centre of gravity is automotive manufacturing and contract logistics.
The shift from demonstration to binding commercial agreement is the defining procurement development of 2026. This is not a transition from pilot to proof-of-concept — it is a transition from proof-of-concept to multi-year contractual commitment, with capital allocations that carry real balance-sheet consequences.
Agility Robotics represents the most publicly documented example of this transition. Its SEC materials, filed in connection with a proposed combination with Churchill Capital Corp XI at a stated pre-money equity value of $2.5 billion (company-claimed), disclose conditional orders exceeding $300 million for the Digit v5 platform (company-claimed). Customer agreements include a verified multi-year RaaS contract with GXO Logistics and a commercial RaaS agreement with Toyota Motor Manufacturing Canada, announced in February 2026 (company-claimed/verified announcement). Apptronik closed a $520 million funding round in February 2026 (verified, Reuters) and in June 2026 launched its Apollo 2 platform alongside a robot-training facility developed with Google DeepMind (verified, Reuters). Xpeng Robotics raised more than $900 million (company-claimed/Reuters) in August 2026, targeting a production rate of 1,000 IRON humanoids per month by year-end 2026 (company-claimed). Boston Dynamics announced that all Atlas deployments for 2026 were committed (company-claimed), including arrangements with Hyundai’s Robotics Manufacturing and Assembly Centre and Google DeepMind.
The primary sectors executing commercial agreements are automotive manufacturing — Mercedes-Benz, BMW, and Schaeffler — and contract logistics, led by GXO Logistics. Industrial manufacturing, with Toyota and Schaeffler as the most visible participants, follows closely. This commercial concentration matters for procurement buyers: the deployment evidence base that informs risk assessment is concentrated in two sectors, which limits cross-sector contractual precedent.
Jonathan Hurst, Co-Founder and Chief Robot Officer of Agility Robotics, stated in connection with the June 2026 transaction announcement (company-claimed executive statement):
“We believe cooperative safety is the critical unlock for scaled humanoid adoption.”
That framing is consistent with what the public commercial evidence confirms: manufacturers are identifying operational architecture — including how robots and human workers share a facility — as a precondition for commercial scalability. Buyers should understand that this commercial objective translates directly into supervision and liability provisions that must be negotiated before deployment.
Platform | Manufacturer | Acquisition Model | Verified Commercial Agreement | Disclosed Economics | Classification |
Digit v5 | Agility Robotics | RaaS / Ownership | GXO Logistics (verified); Toyota (verified announcement) | ~$200K purchase; ~$8.5K/month RaaS; ~$400K / 5-yr ownership total | Company-claimed economics |
Figure 02 | Figure AI | Not publicly specified | BMW deployment (company-claimed metrics) | Not publicly disclosed | Company-claimed deployment data |
Apollo / Apollo 2 | Apptronik | Not publicly specified | Mercedes-Benz testing (verified, Reuters) | Not publicly disclosed | Testing verified |
Atlas | Boston Dynamics | Not publicly specified | 2026 deployments committed (company-claimed) | Not publicly disclosed | Company-claimed |
G1 | Unitree Robotics | Purchase | None confirmed at industrial scale | From ~$13,500 advertised | Company-claimed price |
IRON | Xpeng Robotics | Not yet specified | Commercial sales planned 2027 (company-claimed) | Not publicly disclosed | Company-claimed targets |
Table 1 — Commercial Humanoid Platform Landscape, September 2026
Deployment Evidence: What Commercial Pilots Reveal — and What They Don’t
PHYSICAL AI JOURNAL
Published humanoid deployment evidence confirms that commercial agreements are being signed and that robots are completing operational tasks in industrial settings. What the evidence does not establish is how risk is allocated when deployments fail. No publicly available humanoid deployment announcement has disclosed indemnity terms, liability caps, insurance claim history, SLA penalties, or incident outcomes. Buyers evaluating humanoid contracts in 2026 are making multi-year, multi-hundred-thousand-dollar commitments with no public precedent for how contractual risk has been resolved in practice.
The distinction between deployment announcement and deployment evidence is the most consequential quality control issue in the humanoid procurement market. An announcement confirms that an agreement was signed and that a robot entered a facility. It does not confirm operational continuity, task performance at scale, incident history, or what happens when the deployment fails to meet contractual expectations.
Physical AI Journal applies a six-state deployment classification to distinguish evidence tiers:
Announced → Piloted → Deployed → Operational → Productive → Renewed.
These are not synonyms. A robot that has been deployed may not be operational. A robot that is operational may not be productive at the claimed level. A contract that has been announced may not have proceeded to deployment. An analysis built on deployment announcements that conflates these states overstates the evidence available to buyers.
The specific gap most relevant to procurement is contractual: no publicly available source has disclosed the indemnity structure, liability caps, insurance requirements, SLA definitions, penalty clauses, or termination rights from any commercial humanoid deployment agreement. The deployment evidence circulating in trade and business media is predominantly announcement-based, with operational metrics — uptime, performance at scale, and unit economics — rarely disclosed in publicly available sources. That pattern of non-disclosure extends with equal force to contractual risk data: what a robot achieves in a deployment and what the contract says about failure are two distinct categories of information, and only the former has been selectively disclosed.
Deployment | Agreement Type | Operational Status | Performance Data | Incident Data Public | Contract Terms Public | Classification |
GXO × Agility Digit | Multi-year RaaS (verified) | Commercial deployment confirmed | Not publicly disclosed | Not disclosed | Not disclosed | Verified agreement |
BMW × Figure 02 | Not publicly specified | 11-month deployment (company-claimed) | Company-claimed metrics only | Not disclosed | Not disclosed | Company-claimed |
Mercedes × Apptronik Apollo | Testing arrangement | Testing phase confirmed (verified) | Not publicly disclosed | Not disclosed | Not disclosed | Testing verified |
Schaeffler × Humanoid | Announced commercial agreement | Rollout planned Dec. 2026 (company-claimed) | Not yet available | Not applicable | Not disclosed | Future — company-claimed |
Table 2 — Deployment Evidence Quality Matrix: Four Commercial Agreements. The dominant cell content across the Contract Terms column is Not Disclosed — this is the central finding of this report.
Economics and Risk-Adjusted Deployment Cost
PHYSICAL AI JOURNAL
The cost of deploying a humanoid robot is not its acquisition price. A buyer evaluating Agility’s Digit faces an illustrative five-year ownership cost of approximately $400,000 per robot (company-claimed, Agility SEC materials) — before integration, supervision labour, insurance, potential downtime losses, and any uncapped indemnity exposure. Risk-adjusted deployment cost requires quantifying hardware economics, integration fees, supervision obligations, insurance premiums, SLA penalty exposure, and residual liability after indemnity caps. No publicly available humanoid deployment has disclosed all these inputs. Buyers must require this data from vendors and insurers during procurement.
Acquisition price is the most visible number in a humanoid procurement decision. It is not the most important one. Agility Robotics’ SEC materials present the Digit v5 illustrative five-year ownership economics as follows (all figures company-claimed, management model, not a customer quotation):
● Illustrative hardware purchase price: approximately $200,000 (company-claimed)
● Deployment fee under ownership model: approximately $20,000 (company-claimed)
● Software and maintenance: approximately $36,000 per year (company-claimed)
● Total illustrative five-year ownership cost: approximately $400,000 per robot (company-claimed)
● Alternatively, RaaS subscription: approximately $8,500 per month (company-claimed), representing an illustrative five-year total of approximately $500,000 (company-claimed, as stated in Agility SEC investor materials)
These are management projections for illustrative purposes. They represent one platform at one publicly disclosed commercial stage.
Jörg Burzer, Production Chief at Mercedes-Benz, stated in March 2025 (Reuters, verified):
“The cost will be decisive… when costs reach a two-digit thousand-dollar sum — which is absolutely possible — it will become very interesting.”
That statement confirms that major industrial operators continue to condition scaled deployment on economics reaching a materially lower level than current disclosed figures. Reuters also reported in August 2026 (verified) that Chinese industrial humanoids often carry price points in the range of 300,000–500,000 yuan — a different market context rather than a directly comparable procurement alternative.
Risk-adjusted deployment cost adds the following to hardware economics: integration cost (connecting the robot to existing infrastructure, not disclosed in any deployment); supervision labour cost (public reporting on humanoid deployments confirms contractor or operator supervision is required); insurance premium (no standardised humanoid-specific product with published Western-market pricing exists; buyers must obtain custom quotes); downtime exposure (no publicly available SLA defines this liability); and residual liability after indemnity cap (whatever limit the contract imposes leaves residual exposure with the buyer). The analytical value of this framework is not a single total figure — the data to produce one does not yet exist publicly — but in identifying precisely what buyers must request from the vendor and from the insurer.
Cost Component | Public Data Available | Must Request From Vendor | Must Request From Insurer |
Hardware acquisition / RaaS NPV | Agility illustrative figures (company-claimed) | Firm quotation; volume discounts | — |
Integration cost | Not publicly disclosed | Integration specification and fee | — |
Supervision labour | Not publicly disclosed | Supervision requirement documentation | — |
Insurance premium | Not publicly disclosed | Vendor insurance requirements (minimums) | Custom quote; exclusion schedule |
Downtime / SLA exposure | Not publicly disclosed | SLA definition; penalty structure; remedy | Business-interruption coverage terms |
Residual liability after cap | Not publicly disclosed | Indemnity cap; uncapped categories | Excess-liability options |
Total Risk-Adjusted Cost | Cannot be reliably calculated without vendor and insurer disclosure. Buyers must require all inputs before contract execution. |
|
|
Table 3 — Risk-Adjusted Deployment Cost Framework
Component | Ownership Model | RaaS Model |
Hardware | ~$200,000 purchase (company-claimed) | Included in subscription (company-claimed) |
Deployment fee | ~$20,000 (company-claimed) | Not separately disclosed (company-claimed) |
Software & maintenance (annual) | ~$36,000/yr (company-claimed) | Included in subscription (company-claimed) |
Monthly subscription | N/A | ~$8,500/month (company-claimed) |
Illustrative 5-year total | ~$400,000 (company-claimed) | ~$500,000 (company-claimed — stated figure, Agility SEC investor materials) |
Hardware ownership after contract | Buyer | Agility |
Operational liability (contractually confirmed) | Not publicly disclosed | Not publicly disclosed |
Table 4 — Agility Digit: Illustrative Deployment Economics by Model. All figures company-claimed, Agility SEC materials, 2026.
Named Deployment Case Studies: Risk Intelligence From Four Commercial Agreements
PHYSICAL AI JOURNAL
Four commercial humanoid deployments are publicly documented with sufficient evidence to analyse: GXO Logistics × Agility Digit (verified RaaS agreement, Georgia, USA); BMW × Figure AI (company-claimed deployment metrics, Spartanburg, South Carolina); Mercedes-Benz × Apptronik Apollo (verified testing phase, Berlin and Kecskémét, Hungary); and Schaeffler × Humanoid (announced future commercial agreement, Germany). In every case, contractual risk-allocation terms — indemnity caps, insurance requirements, liability attribution, SLA penalties, and incident outcomes — are not publicly disclosed.
Case Study 1 — GXO Logistics × Agility Digit: The Verified RaaS Precedent
GXO Logistics signed a multi-year RaaS agreement with Agility Robotics in June 2024, deploying Digit at its SPANX fulfilment facility in Flowery Branch, Georgia, USA. The deployment followed a proof-of-concept pilot conducted in late 2023. GXO described the arrangement as an industry-first formal commercial humanoid deployment. The operational task was tote handling: moving totes from autonomous mobile robots and placing them onto conveyors. Evidence classification: verified customer and company announcement.
This case represents the strongest publicly documented example of a humanoid procurement transition from pilot to multi-year commercial contract. Its procurement intelligence value is not in what was disclosed, but in what was not. No source — customer, manufacturer, or trade press — has published the indemnity structure, liability caps, insurance requirements, SLA definitions, incident records, or uptime performance data from this agreement. Under a RaaS structure, hardware ownership sits with Agility. Operational liability allocation — the question buyers most need answered — was not publicly stated.
Buyer implication: This case confirms that major contract logistics operators are entering multi-year humanoid RaaS agreements. It does not establish a risk-allocation template. Buyers should not treat GXO’s entry as evidence that standard RaaS terms provide adequate risk transfer.
Case Study 2 — BMW × Figure AI: Company-Claimed Performance, No Risk Data
BMW Group deployed Figure AI’s Figure 02 platform at its Spartanburg, South Carolina assembly facility. Figure AI has stated that the deployment ran for 11 months (company-claimed), with the robot operating on 10-hour, Monday-to-Friday shifts (company-claimed). Figure AI claims the robot loaded more than 90,000 parts (company-claimed), accumulated more than 1,250 operating hours (company-claimed), and contributed to production of more than 30,000 BMW X3 vehicles (company-claimed). These metrics have not been independently audited. Evidence classification: company-claimed performance data; deployment confirmed.
The productivity claims, if substantiated, would represent the most operationally extensive publicly documented humanoid deployment in automotive manufacturing. The critical procurement limitation is that none of these metrics speaks to the contractual risk structure. No source has published what liability Figure AI accepted or excluded in its BMW agreement; what insurance either party carried; what SLA governed uptime; or what remedy BMW would have had if the robot caused injury or equipment damage.
Buyer implication: Headline operational metrics from manufacturer announcements do not substitute for contractual risk data. A buyer cannot assess exposure from a press release. Independent contractual confirmation is required for each risk category in Table 6.
Case Study 3 — Mercedes-Benz × Apptronik Apollo: Economics Drive the Decision
Mercedes-Benz began testing Apptronik’s Apollo humanoid at its Digital Factory Campus in Berlin and at its facility in Kecskémét, Hungary, with Reuters reporting confirmed in March 2025. A small number of Apollo robots were being trained through teleoperation for specific manufacturing tasks, including moving components to production lines and quality inspection functions. Evidence classification: verified, Reuters reporting based on Mercedes disclosures.
Production Chief Burzer’s assessment — cited in full in Section 5 above — identified cost as the primary condition for scaled deployment: the figure must reach a two-digit thousand-dollar sum before broader commitment becomes viable (Reuters, March 2025, verified). For procurement buyers, this provides a useful analytical anchor: a major industrial operator with engineering resources and financial scale is applying a cost threshold before committing to commercial deployment.
Buyer implication: The cost-threshold approach used by Mercedes-Benz is analytically sound but incomplete. Economics are the necessary condition for deployment; contractual risk allocation is the sufficient condition. A favourable acquisition price does not resolve indemnity, insurance, or termination risk.
Case Study 4 — Schaeffler × Humanoid: The Pre-Deployment Contract Design Moment
Schaeffler, the German automotive components manufacturer, has announced a commercial agreement with Humanoid for an initial rollout scheduled between December 2026 and June 2027 (company-claimed, Reuters, May 2026). The announced scale target is 1,000–2,000 robots across Schaeffler’s global manufacturing sites by 2032 (company-claimed). Initial deployment locations are Herzogenaurach and Schweinfurt, Germany. Applications announced include box handling and near-full-scale operational testing. Evidence classification: announced commercial agreement; no completed deployment to evaluate.
This case is the most instructive for this report’s central argument precisely because no deployment has yet occurred. The moment between signed agreement and first robot activation is the window this report is designed to serve. The contractual risk allocation — who carries liability for bodily injury, property damage, software failure, and business interruption — must be established before the first robot enters the facility.
Buyer implication: Announced large-scale agreements with long deployment timelines create the clearest opportunity to negotiate comprehensive risk allocation before operations begin. That opportunity closes once deployment is underway.
Deployment | Operational Status | Contractual Terms Public | Incident Data Public | Insurance Data Public | SLA Data Public |
GXO × Digit | Commercial RaaS agreement verified | Not disclosed | Not disclosed | Not disclosed | Not disclosed |
BMW × Figure 02 | 11-month deployment (company-claimed) | Not disclosed | Not disclosed | Not disclosed | Not disclosed |
Mercedes × Apollo | Testing phase verified | Not disclosed | Not disclosed | Not disclosed | Not disclosed |
Schaeffler × Humanoid | Pre-deployment (rollout Dec. 2026) | Not disclosed | Not applicable | Not disclosed | Not disclosed |
Table 5 — Case Study Risk Intelligence Summary
The Humanoid Buyer Risk Allocation Matrix
PHYSICAL AI JOURNAL
Industrial buyers deploying humanoids must negotiate contractual assignment of ten risk categories before signing: hardware defects, AI and software failures, post-deployment software update impacts, third-party integration failures, supervision and operator obligations, bodily injury liability, property and equipment damage, business interruption and SLA performance, cyber incident exposure, and vendor failure or platform discontinuation. For each category, buyers must obtain written contractual allocation — including indemnification provisions, liability caps, insurance requirements, and termination rights — before committing capital. No publicly available humanoid deployment contract has disclosed terms across all ten categories.
The humanoid robot risk allocation challenge is not legal in origin — it is commercial. The question is not which party a court would hold responsible after an incident. The question is which party has agreed, in writing, before the contract is signed, to bear each category of financial exposure. That agreement is what the matrix below is designed to produce.
Physical AI Journal has identified ten categories that every humanoid deployment contract must address. A critical instruction for using this matrix: the responsible party column reflects what the contract must establish — not what any specific manufacturer currently accepts or excludes. Where no public evidence exists, the matrix states: No Public Precedent — contractual confirmation required.
# | Risk Category | Who Can Originate It | Contractual Provision Required | Buyer Action | Public Precedent |
1 | Hardware defect | Manufacturer | Product warranty; defect indemnification; replacement obligation; liability cap | Obtain written warranty scope and duration; confirm replacement timeline | No public contract terms disclosed |
2 | AI / software failure | Manufacturer; AI software provider | Software defect indemnification; AI-decision liability clause; separation of hardware and software liability | Confirm which party indemnifies AI-decision failures; confirm whether software provider is a separate party to the contract | No public contract terms disclosed |
3 | Software update impact | Manufacturer (remote update) | Update-notification obligation; pre-deployment testing requirement; liability for behaviour changes post-update | Confirm right to approve or defer updates; confirm who carries liability if update causes incident | No public contract terms disclosed |
4 | Third-party integration failure | System integrator; manufacturer | Integrator indemnification; manufacturer liability boundary at integration interface; change-control clause | Confirm where manufacturer liability ends and integrator liability begins; include integrator as party to indemnity structure | No public contract terms disclosed |
5 | Supervision / operator failure | Site operator; RaaS provider; manufacturer | Supervision obligation clause; staffing requirements; liability allocation for supervisor absence or inadequacy | Confirm who is contractually responsible for supervision; confirm staffing level required by contract | WSJ reporting confirms contractor supervision in use at a humanoid deployment site; contractual terms not public |
6 | Bodily injury | All parties (depending on failure origin) | Bodily injury indemnification; liability cap (or uncapped confirmation); employer-liability insurance coordination; third-party injury clause | Confirm bodily injury indemnity is explicit; confirm whether cap applies; confirm insurance coordination with existing employer liability policy | No public contract terms disclosed |
7 | Property and equipment damage | Manufacturer; integrator; operator | Property damage indemnification; equipment damage liability clause; exclusions schedule | Confirm damage to production equipment, product in process, and facility is covered; confirm exclusions | No public contract terms disclosed |
8 | Business interruption and downtime | Manufacturer; integrator; RaaS provider | SLA definition; uptime guarantee; downtime remedy; SLA credit mechanism; business-interruption compensation clause | Define uptime standard; confirm penalty structure; confirm credit calculation methodology | No public SLA terms disclosed for any humanoid deployment |
9 | Cyber incident | Manufacturer (remote access); operator; integrator | Cyber incident liability clause; remote-access security obligation; data-breach indemnification; malware/hijacking coverage | Confirm manufacturer’s remote-access security obligations; confirm cyber insurance applies to humanoid-related incidents | Munich Re has identified remote hijacking as an emerging physical-AI exposure; no public contract terms disclosed |
10 | Vendor failure / platform discontinuation | Manufacturer | Software escrow clause; spare-parts obligation; platform-continuity commitment; termination rights on vendor insolvency or acquisition | Confirm termination rights if manufacturer fails; confirm software escrow arrangement; confirm spare-parts obligations | No public contract terms disclosed; Agility SPAC transaction changes counterparty structure mid-cycle |
Table 6 — Humanoid Buyer Risk Allocation Matrix (Core Deliverable). The Responsible Party column reflects what contracts must establish, not what any manufacturer currently accepts. Where no evidence exists: No Public Precedent — contractual confirmation required.
Two categories warrant particular attention in 2026. Category 10 (vendor failure and platform discontinuation) has become more materially relevant following Agility Robotics’ announced combination with a SPAC vehicle at a stated pre-money equity value of $2.5 billion (company-claimed), with more than $620 million in expected gross transaction proceeds (company-claimed). A transaction of this nature changes the counterparty structure of existing multi-year agreements.
Category 3 (software update impact) carries specific urgency absent from conventional equipment procurement. A humanoid robot can receive an AI model update, firmware revision, or behaviour-policy change remotely, without physical intervention. If that update alters the robot’s operational behaviour and an incident results, the failure origin is the updated software — not the hardware configuration the buyer accepted at deployment. No publicly available humanoid contract specifies how liability is allocated in this scenario.
RaaS vs Outright Purchase: Does the Deployment Model Change Your Risk Exposure?
PHYSICAL AI JOURNAL
Robots-as-a-Service (RaaS) agreements transfer hardware ownership to the vendor and typically include maintenance and replacement provisions. RaaS does not automatically transfer operational liability to the vendor. Bodily injury risk, supervision responsibility, business interruption exposure, and data obligations may remain with the site operator regardless of ownership model. Whether a specific RaaS agreement shifts meaningful liability to the vendor depends entirely on the contract’s indemnification, insurance, and supervision provisions — which no publicly available humanoid RaaS agreement has disclosed. Buyers must not treat RaaS as a risk-transfer mechanism without contractual confirmation.
RaaS shifts hardware ownership to the vendor; it transfers operational liability only to the extent the contract explicitly establishes. The GXO × Agility RaaS agreement — the only verified large-scale humanoid RaaS commercial contract in the public evidence base — has not disclosed its indemnification structure. That absence means buyers cannot use this agreement as evidence that RaaS is, in practice, a risk-transfer mechanism.
What RaaS changes structurally is straightforward: the manufacturer retains ownership of the hardware throughout the contract term. This has two procurement implications. First, the manufacturer bears the asset-ownership risk — depreciation, residual value, and platform obsolescence sit with the vendor. Second, replacement and maintenance obligations are typically included in the subscription, removing an operational risk category from the buyer’s scope. These are meaningful commercial advantages.
What RaaS does not change, unless the contract explicitly states otherwise, is operational liability.
A robot that causes bodily injury, property damage, or a production shutdown at the buyer’s facility creates a financial loss. Whether the manufacturer’s ownership of the hardware means the manufacturer indemnifies that loss is a contractual question — not a structural consequence of the ownership model. In jurisdictions where employer liability sits with the operator of a workplace, a RaaS arrangement that does not explicitly address this may leave the buyer as the primary liable party for incidents involving its employees, regardless of who owns the robot.
Risk Category | Ownership Model | RaaS Model | Contractual Confirmation Required |
Hardware asset risk | Buyer bears depreciation and residual value | Vendor bears asset risk | Confirm scope of replacement obligation |
Maintenance obligation | Contract-dependent; typically buyer | Typically included in subscription | Confirm coverage scope and response times |
Bodily injury liability | Not automatically assigned by ownership | Not automatically transferred by RaaS | Explicit indemnification clause required — both models |
Supervision responsibility | Not defined by ownership model | Not defined by ownership model | Explicit supervision obligation clause required |
Business interruption | Not automatically assigned | Not automatically assigned | SLA with downtime remedy required — both models |
Software update liability | Not automatically assigned | Not automatically assigned; vendor retains update capability | Update-notification and liability clause required — both models |
Table 7 — RaaS vs Ownership: Risk Allocation Comparison
The decision between RaaS and outright ownership should be made on financial and strategic grounds: capital efficiency, balance-sheet treatment, and operational flexibility. The risk allocation decision is separate — it must be made on contractual grounds, regardless of which acquisition model the buyer chooses.
Competitive Platform Risk Procurement Comparison
PHYSICAL AI JOURNAL
A risk-informed humanoid platform comparison cannot be conducted on acquisition price alone. Agility Digit offers the most publicly documented commercial economics and the only verified multi-year RaaS agreements at scale. Figure 02 has a verified BMW deployment but limited public contract data. Apptronik Apollo has verified automotive testing and substantial institutional backing. Boston Dynamics Atlas carries Hyundai Group financial support. Unitree G1 presents a substantially different deployment profile at its advertised ~$13,500 entry price (company-claimed). Xpeng IRON targets high-volume production from 2027. All specifications and pricing are company-claimed unless separately noted.
Six commercial humanoid platforms are active in the industrial market in 2026, each presenting a distinct risk procurement profile that acquisition price alone does not capture. This section compares them on procurement-relevant risk dimensions — acquisition model availability, verified commercial deployment history, manufacturer financial position as a counterparty-risk indicator, software and AI dependency, and publicly disclosed warranty or support terms. It does not rank platforms on technical specifications, operational performance, or suitability for any specific deployment environment.
The price range across these platforms spans two orders of magnitude: Unitree’s G1 carries an advertised starting price of approximately $13,500 (company-claimed), while Agility’s five-year illustrative ownership cost sits at approximately $400,000 (company-claimed). These figures represent fundamentally different product categories — different deployment environments, commercial support structures, AI integration depth, and expected supervision requirements. Comparing them on acquisition price produces a misleading procurement picture. The relevant comparison is on risk-adjusted procurement dimensions, as presented below.
Jeff Cardenas, CEO and Co-Founder of Apptronik, stated in February 2025 (Reuters, verified):
“This represents an inflection point for the industry.”
That framing is consistent with the public commercial evidence: capital commitments are active and platforms are entering industrial environments across automotive manufacturing and logistics. The counterparty risk column in Table 8 reflects a complementary reality: most humanoid manufacturers remain pre-revenue at scale, with valuations that carry significant assumptions about future commercial performance.
Platform | Manufacturer | Acquisition Model | Verified Commercial Deployment | Manufacturer Financial Position | AI / Software Dependency | Warranty / Support Terms | Counterparty Risk Indicator |
Digit v5 | Agility Robotics | RaaS / Ownership | Yes — GXO, Toyota (verified) | SPAC combination announced; >$620M gross proceeds expected (company-claimed) | High — continuous AI/ML development | Not publicly disclosed | Capital transition underway; monitor post-transaction structure |
Figure 02 | Figure AI | Not publicly specified | BMW deployment (company-claimed metrics) | >$1B committed (company-claimed); $39B post-money (company-claimed) | High — AI-native platform | Not publicly disclosed | Pre-revenue at scale; high valuation relative to disclosed revenue |
Apollo 2 | Apptronik | Not publicly specified | Mercedes-Benz testing (verified); no scaled production deployment confirmed | $520M funding (verified, Reuters); ~$5B valuation (company-claimed) | High — Google DeepMind collaboration | Not publicly disclosed | Well-capitalised; institutional backing; pre-scaled deployment |
Atlas | Boston Dynamics | Not publicly specified | 2026 deployments committed (company-claimed); scaled deployment not yet confirmed (Reuters) | Hyundai Group ownership; 30,000-robot annual capacity target by 2028 (company-claimed) | High — AI-native; remote operation capability | Not publicly disclosed | Parent-group backing reduces insolvency risk; platform still pre-scaled |
G1 | Unitree Robotics | Purchase | None confirmed at industrial scale | Public-market debut experienced regulatory scrutiny (Reuters) | Configurable; AI integration varies | Not publicly disclosed | Volatile public-market position; different risk profile from enterprise vendors |
IRON | Xpeng Robotics | Not yet specified | None at production scale; commercial sales planned 2027 | >$900M funding (company-claimed/Reuters); >$6.3B valuation (company-claimed/Reuters) | High — AI-native; Chinese market context | Not publicly disclosed | Well-funded; pre-deployment at scale; different regulatory environment |
Table 8 — Humanoid Platform Risk Procurement Comparison. All specifications and pricing are company-claimed unless separately noted as verified.
Friction, Risk and Unresolved Issues
PHYSICAL AI JOURNAL
The six most commercially significant unresolved issues in humanoid deployment contracts are: undefined supervision responsibility; absent software-update liability frameworks; immature insurance products without historical loss data; regulatory transition risk from the EU AI Act’s August 2028 high-risk machinery deadline; vendor financial durability as a counterparty risk in multi-year agreements; and undefined ownership of incident logs and operational data required to attribute fault. Each must be addressed contractually before deployment — none has been resolved by publicly available industry practice or standard contract terms.
Friction 1 — Supervision Responsibility Is Contractually Undefined
Public reporting on Schaeffler’s Agility Digit deployment has confirmed that the robot operated separately from human workers and that a contractor was stationed nearby to monitor it during operation (VERIFIED — WSJ reporting). This supervision configuration implies two costs: the contractor’s time and the associated liability if supervision is inadequate. Who is contractually responsible for maintaining adequate supervision? Is it the manufacturer, the integrator, the RaaS provider, or the site operator? No publicly available humanoid deployment contract has addressed this question.
Friction 2 — Software-Update Liability Has No Established Model
A humanoid platform can receive updates to its firmware, AI perception model, behaviour policy, or decision logic remotely, between operational shifts, without physical intervention by the buyer. If that behavioural change — not a hardware failure but a software update issued unilaterally by the manufacturer — causes an incident after deployment, the contractual question is: who carries the liability? The EU’s Product Liability Directive, which entered into force in December 2024 (VERIFIED — European Commission), explicitly covers software, AI systems, and product-related digital services. This regulatory framework is relevant context, but it does not substitute for contractual allocation.
Friction 3 — Insurance Products Are Evolving Without Established Loss History
Reuters’ May 2026 analysis (VERIFIED) confirmed that traditional commercial general liability and business-interruption insurance can remain applicable to physical AI risks — but that insurers are reviewing exclusions and developing AI-specific approaches. Munich Re (VERIFIED) has identified bodily injury, property damage, production shutdown, remote hijacking, and malware as emerging physical-AI exposures. No standardised humanoid-specific insurance product with published Western-market coverage terms, premium benchmarks, or exclusion schedules currently exists. Buyers must obtain custom insurance assessments before deployment — and must confirm that existing CGL and employer-liability policies cover humanoid-related incidents without exclusion.
Friction 4 — Regulatory Transition Risk Sits Inside Multi-Year Contracts
The European Commission has confirmed (VERIFIED) that high-risk AI systems embedded in physical products — including machinery — carry an application date of 2 August 2028 under the EU AI Act, following the AI Omnibus which entered into force on 31 July 2026. For buyers in the EU signing three- or five-year humanoid deployment agreements in 2026 or 2027, this compliance date sits within the contract term.
If compliance requires modifications to the platform, the AI system, or the deployment configuration, who bears the cost? This unresolved procurement question belongs in the regulatory-transition clause of every EU-jurisdiction deployment contract signed before August 2028.
Friction 5 — Vendor Financial Durability Is a Counterparty Risk, Not a Technology Risk
Most humanoid manufacturers are pre-revenue at commercial scale. Agility Robotics’ announced SPAC combination (company-claimed pre-money equity value: $2.5 billion; expected gross transaction proceeds: more than $620 million (company-claimed)) represents a capital-structure change that affects existing and future multi-year counterparties. Reuters reported in August and September 2026 (VERIFIED) that Chinese humanoid manufacturers are facing scrutiny over whether government-supported production translates into recurring commercial revenue. A buyer signing a five-year agreement with a manufacturer that subsequently fails, is acquired, or exits the humanoid market may find that spare parts, software support, and contractual remedies are not available.
Friction 6 — Incident Log Ownership Is Undefined and Consequential
After a humanoid deployment incident, fault attribution requires access to operational data: the robot’s sensor logs at the time of the event, the AI decision record, the software version in operation, the maintenance log, and any remote access or update activity that preceded the incident.
Who owns this data?
Where is it stored?
Who has the right to access it during a dispute?
No publicly available humanoid deployment agreement has addressed these questions. Buyers seeking to establish that a failure originated with the manufacturer — rather than with site operation — may lack access to the data needed to make that case.
Strategic Recommendations: The 10-Point Pre-Signature Buyer Gate
PHYSICAL AI JOURNAL
Before signing a humanoid deployment contract, industrial buyers must confirm ten provisions in writing: bodily injury liability allocation, property and equipment damage responsibility, SLA and business interruption terms, software and AI failure indemnification, integration failure liability, cyber incident coverage, right to access incident logs, regulatory transition cost allocation, vendor-failure contractual remedies, and minimum insurance requirements from all parties. A contract that does not address all ten points leaves material financial exposure unassigned. Buyers must not proceed to scaled deployment without independent legal review and confirmed insurance coverage.
The ten provisions below represent the minimum contractual standard for an industrial humanoid deployment. They are framed as buyer requirements — items that must be confirmed in writing before any deployment agreement is executed. This is not a legal compliance checklist. It is a commercial risk-allocation framework, and buyers should obtain independent legal counsel to review the specific contract language for each item before signing.
1. Bodily Injury Liability
The contract must explicitly state which party indemnifies the buyer for bodily injury claims arising from the robot’s operation. It must specify whether a liability cap applies to bodily injury claims (or confirm that bodily injury indemnification is uncapped), and must coordinate with the buyer’s existing employer-liability and CGL insurance. Bodily injury is the highest-severity risk category and the provision where contractual ambiguity carries the greatest financial consequence.
2. Property and Equipment Damage
The contract must specify indemnification for damage caused by the robot to the buyer’s facility, production equipment, and in-process product. It must define whether physical impact, software-driven operational errors, and consequential damage to adjacent equipment are all covered. The exclusions schedule requires particular attention.
3. Business Interruption and SLA Performance
The contract must define the uptime standard the robot is required to achieve, the measurement methodology, the consequence of underperformance, and the credit or compensation mechanism. Business interruption costs in an industrial production environment can exceed the robot’s acquisition cost within days. An SLA without a defined remedy is not an SLA.
4. Software and AI Failure Indemnification
The contract must separately address AI-decision failures and software failures — including failures that occur after a manufacturer-issued update. It must specify which party indemnifies losses caused by an AI perception error, a model decision that causes an incident, or a behaviour change introduced by a software update. This provision cannot be satisfied by a general product warranty that addresses only hardware defects.
5. Integration Failure Liability
The contract must define where the manufacturer’s liability ends and the integrator’s liability begins. If a failure occurs at the interface between the robot and the buyer’s existing production systems, and that interface was configured by a third-party integrator, both the manufacturer and the integrator must be parties to the indemnification structure.
6. Cyber Incident Clause
The contract must address remote access compromise, malware, and unauthorised hijacking as distinct risk categories. It must specify the manufacturer’s security obligations for remote access systems and define liability for cyber-incident consequences. Munich Re has identified remote hijacking as an emerging physical-AI exposure. Confirming that the buyer’s cyber insurance applies to humanoid-related incidents requires insurer confirmation, not assumption.
7. Incident Data and Log Access
The contract must establish the buyer’s right to access all operational data, sensor logs, AI decision records, and software version histories required to attribute fault after an incident. This includes data held on the manufacturer’s systems. The right must be available immediately after an incident, not subject to manufacturer consent during a dispute.
8. Regulatory Compliance Transition
For any deployment in the EU or UK jurisdiction that extends beyond August 2028, the contract must address who bears the cost of compliance modifications required by the EU AI Act’s high-risk machinery provisions. If the manufacturer is responsible for delivering a compliant platform, this must be stated. If the buyer assumes compliance risk, this must be a deliberate contractual decision, not an oversight.
9. Vendor Failure and Platform Discontinuation
The contract must establish what happens if the manufacturer ceases to support the platform, is acquired, enters insolvency, or discontinues the product line during the contract term. Provisions should include software escrow arrangements, spare-parts obligations for a defined period after platform discontinuation, and buyer termination rights without penalty if the manufacturer fails to maintain support obligations.
10. Insurance Requirements
The contract must specify minimum insurance coverage levels required of the manufacturer, the integrator, and the buyer. Evidence of coverage — certificates of insurance or equivalent — should be required from all parties before deployment commences. The buyer’s broker must confirm that existing policies apply to humanoid-related incidents without exclusion.
# | Provision | What It Must State | Why It Matters | Deploy Without It? |
1 | Bodily injury liability | Which party indemnifies; cap or uncapped; insurance coordination | Highest-severity single exposure in a humanoid deployment | No — hard requirement |
2 | Property and equipment damage | Scope of damage covered; exclusions schedule | Production equipment damage can exceed robot cost | No — hard requirement |
3 | Business interruption / SLA | Uptime standard; measurement; remedy; credit mechanism | Production stoppage costs accumulate rapidly | No — escalate to legal review |
4 | Software / AI failure indemnification | Separate clause for AI-decision and update-driven failures | Not covered by standard hardware warranty | No — hard requirement |
5 | Integration failure liability | Boundary between manufacturer and integrator liability | Gap between contracts leaves buyer exposed | No — escalate to legal review |
6 | Cyber incident clause | Remote-access security obligations; incident liability; insurance confirmation | Remote hijacking identified as emerging exposure (Munich Re) | No — escalate to legal review |
7 | Incident log access | Immediate right to all operational data post-incident | Without data access, fault attribution is impossible | No — escalate to legal review |
8 | Regulatory transition | Who bears compliance cost for EU AI Act — August 2028 | Multi-year contracts span the compliance date | No — required for EU jurisdiction deployments |
9 | Vendor failure remedies | Software escrow; spare-parts obligation; termination rights | Most manufacturers are pre-revenue at scale | No — hard requirement |
10 | Insurance requirements | Minimum coverage for all parties; evidence before deployment | Unconfirmed coverage leaves gaps at point of loss | No — hard requirement |
Table 9 — 10-Point Pre-Signature Buyer Gate
The Deploy / Pilot / Wait Decision Framework
✓ DEPLOY All 10 provisions contractually addressed. Independent legal review completed. Insurer has confirmed coverage. All parties have provided evidence of insurance. | ● PILOT 7–9 provisions addressed. Proceed with a time-limited, scope-limited pilot. Remaining provisions must be resolved before scaled deployment is approved. | ✗ WAIT Fewer than 7 provisions addressed — or bodily injury, software-update liability, or vendor-failure provisions are absent. Do not proceed to any deployment. |
A Procurement Buyer who has worked through this gate can enter any humanoid vendor negotiation with a specific, written list of the ten provisions that must be confirmed before capital is committed — and can classify any proposed agreement as Deploy-ready, Pilot-eligible, or not yet signable without relying on vendor assurance alone.
Executive FAQ
PHYSICAL AI JOURNAL
An Automation/Procurement Buyer evaluating a humanoid deployment contract in 2026 faces six questions that no publicly available commercial contract has answered: who bears bodily injury liability; what to negotiate before signing; whether RaaS transfers risk; what insurance is required; what happens if the manufacturer fails during a five-year commitment; and who is responsible for supervision. Each answer below is evidence-grounded, reflecting the state of public contractual disclosure as of September 2026. Where no contract evidence exists, this is stated directly.
Q1. Who is liable if a humanoid robot injures a worker in my facility?
Liability for worker injury in a humanoid deployment depends on the contractual terms between the buyer, manufacturer, integrator, and RaaS provider — not on a universal legal rule. No publicly available humanoid deployment contract discloses how bodily injury liability is allocated. Buyers must obtain explicit written indemnification provisions before deployment commences, and must confirm that existing employer-liability and CGL insurance applies to humanoid-related incidents without exclusion. Independent legal counsel is required before signing any agreement that does not explicitly address this exposure.
Q2. What should companies negotiate in a humanoid robot deployment contract?
Ten provisions must be addressed in writing before any humanoid deployment agreement is signed: bodily injury liability, property damage, SLA and downtime, software and AI failure indemnification, integration fault allocation, cyber incident coverage, incident log access rights, regulatory transition cost allocation (particularly EU AI Act — August 2028), vendor-failure remedies, and minimum insurance requirements. Any contract that does not address all ten leaves material financial exposure unallocated between the parties. Refer to Table 9 for the complete pre-signature gate.
Q3. Does RaaS transfer liability from the buyer to the humanoid robot manufacturer?
RaaS transfers hardware ownership to the vendor and typically includes maintenance and replacement provisions. It does not automatically transfer operational liability. Bodily injury risk, supervision obligations, and business interruption exposure may remain with the site operator regardless of whether the robot is leased or purchased. Any RaaS contract should be evaluated against the same ten provisions required of an outright purchase agreement — the ownership model does not remove the risk categories that must be contractually assigned.
Q4. What insurance does a company need before deploying a humanoid robot?
Standard commercial general liability and business interruption insurance are the foundation, but their coverage of humanoid-specific incidents — including AI decision failures, software-update-induced events, and cyber incidents — must be confirmed with the insurer before deployment. Buyers should additionally require the manufacturer and integrator to maintain specified minimum insurance levels and provide evidence of coverage before deployment commences. No standardised humanoid-specific insurance product with published Western-market coverage terms currently exists. Munich Re has identified bodily injury, property damage, production shutdown, and remote hijacking as emerging physical-AI exposures.
Q5. What happens if the humanoid robot manufacturer goes out of business during a five-year contract?
No publicly available humanoid deployment contract establishes what happens to the buyer’s investment, the deployed robots, software support, and spare parts if the manufacturer fails during the contract term. Buyers must negotiate vendor-failure contractual remedies — including software escrow, spare-parts obligations, and termination rights — before signing any multi-year agreement. Agility Robotics’ announced SPAC combination (company-claimed pre-money equity value: $2.5 billion) illustrates that capital-structure changes can occur within a multi-year contract term. The counterparty risk analysis should be part of every procurement decision.
Q6. Who is responsible for supervision when a humanoid robot is operating near human workers?
Supervision responsibility in humanoid deployments has not been standardised by contract or regulation. Public reporting on Schaeffler’s Digit deployment confirms that the robot operated separately from human workers and that a contractor was stationed nearby to monitor it during operation (VERIFIED — WSJ reporting). Whether that supervision obligation sits with the manufacturer, integrator, RaaS provider, or site operator — and who carries liability if supervision is inadequate — must be established contractually before deployment. The cost of the supervision arrangement must also be included in the full economic assessment.
Scope and Disclaimer
Physical AI Journal, operated by Sekason Research Limited, United Kingdom (Company Registration No. 14339910), provides market intelligence, commercial analysis, and procurement framework guidance only.
Nothing in this report constitutes legal advice, financial advice, investment advice, tax advice, engineering advice, or safety-certification opinion of any kind. Physical AI Journal does not evaluate, test, certify, or endorse the safety, technical conformity, regulatory compliance, or fitness-for-purpose of any humanoid robot platform, deployment configuration, or robotic system described in this report.
Readers must obtain independent legal counsel, insurance advice, engineering assessment, and regulatory guidance before making procurement, deployment, contractual, or investment decisions. Platform specifications, pricing, deployment status, and commercial arrangements may change without notice. Company-claimed figures originate from manufacturers or vendors and have not been independently verified by Sekason Research Limited.
This report is published under the editorial standards and research methodology of Physical AI Journal Contact: contact@sekasonresearch.com.
References and Strategic Sources
This report is backed by authoritative research, independent verification, and structured analytical methodology, drawing from 23 strategic sources across regulatory bodies, corporate filings, robotics manufacturers, financial news, and risk management firms:
European Commission – EU Adapts Product Liability Rules to Digital Age (9 Dec 2024, ec.europa.eu) — Verified — Tier 1
European Commission – AI Act / Regulatory Framework (Updated 2026, digital-strategy.ec.europa.eu) — Verified — Tier 1
European Commission – AI Omnibus Enters Into Force (31 Jul 2026, ec.europa.eu) — Verified — Tier 1
SEC / Agility Robotics – Agility–Churchill Capital Corp XI Transaction Materials (2026, sec.gov) — Company-claimed — Tier 1
SEC / Agility Robotics – S-4 Registration Statement (4 Sep 2026, sec.gov) — Company-claimed — Tier 1
Agility Robotics – Toyota Commercial RaaS Agreement Announcement (19 Feb 2026, agilityrobotics.com) — Company-claimed — Tier 1
GXO Logistics – Industry-First Multi-Year Agility Agreement (27 Jun 2024, gxo.com) — Verified — Tier 1
Figure AI – Figure 02 BMW Deployment Results (19 Nov 2025, figure.ai) — Company-claimed — Tier 1
Boston Dynamics – Atlas Product Announcement (5 Jan 2026, bostondynamics.com) — Company-claimed — Tier 1
Boston Dynamics – Atlas Product Specifications (Current 2026, bostondynamics.com) — Company-claimed — Tier 1
Unitree Robotics – G1 Product Specifications / Pricing (Current 2026, unitree.com) — Company-claimed — Tier 1
Reuters – Apptronik Raises $520M (11 Feb 2026, reuters.com) — Verified / Company-claimed data — Tier 2
Reuters – Mercedes-Benz Tests Apptronik Apollo (18 Mar 2025, reuters.com) — Verified — Tier 2
Reuters – Humanoid–Schaeffler Deployment (13 May 2026, reuters.com) — Company-claimed — Tier 2
Reuters – Apptronik Apollo 2 Launch (30 Jun 2026, reuters.com) — Verified / Company-claimed — Tier 2
Reuters – Xpeng Robotics Funding Round (24 Aug 2026, reuters.com) — Verified / Company-claimed — Tier 2
Reuters – China Humanoid Commercial Viability (27 Aug 2026, reuters.com) — Verified reporting — Tier 2
Reuters – China Humanoid IPO Scrutiny (9 Sep 2026, reuters.com) — Partially verified — Tier 2
Reuters – Boston Dynamics / Atlas Commercial Outlook (14 Sep 2026, reuters.com) — Verified reporting — Tier 2
Reuters Legal Analysis – Physical AI Insurance and Liability (13 May 2026, reuters.com) — Verified — Tier 2
Munich Re – Cyber Insurance Risks & Physical AI (2026, munichre.com) — Verified — Tier 1
Marsh – 2026 Litigation Trends & Casualty Risks (2026, marsh.com) — Verified — Tier 1
The Wall Street Journal – Reporting on Schaeffler’s Agility Digit Deployment and Supervision Configuration (2025–2026, wsj.com) — Verified — trade press reporting; cited in verified research data Section F
© 2026 Sekason Research Limited · Physical AI Journal · physicalaijournal.org · Strategic Intelligence for the Robotics Decade · Proprietary and Confidential Company Registration No. 14339910 · London, United Kingdom · contact@sekasonresearch.com
