Can a Humanoid Working Beside People Be Made Safe? Figure AI Meets Its Binding Constraint

A red padlock secured through a lockout hasp on a circuit breaker in a grey electrical panel. A red and white tag reading "Danger, Do Not Operate" hangs from the hasp, and a second label reads "Danger, Locked Out." — Cut the power, lock it, tag it. The standard assumes the machine is now safe.Source: © Brady Worldwide, Inc.

Executive Summary

In June 2026, BMW Group published a statement claiming that its 11-month deployment of Figure 02 at Plant Spartanburg “proved that humanoids are no longer lab experiments — they can be a valuable asset in establishing a flexible, reliable manufacturing workforce.” The claim is worth examining carefully, because Figure AI has since been valued at $39 billion, is shipping robots at one per hour from its California factory, and is preparing to enter European markets governed by a mandatory new machinery regulation taking effect in January 2027. The question this article addresses is whether a dynamically-stable humanoid working alongside people can be built to a defensible safety case under ISO 10218:2025, OSHA lockout-tagout requirements, and the EU Machinery Regulation 2023/1230. The answer, based on available evidence, is undetermined, and the uncertainty is concentrated in a single structural gap that no amount of production throughput resolves.

Figure AI's valuation rose 78-fold in 28 months, reaching $39 billion on near-zero disclosed revenue

Post-money valuations at each primary round from Series A (May 2023) through Series C (September 2025); secondary market Forge price as of August 2026 shown at ~$34.2B.

The Claim, Examined

The BMW statement is specific in what it asserts and careful in what it does not. “Valuable asset” and “flexible, reliable” are operational claims about task performance in a live factory environment. The statistics behind them are impressive on their face: over 90,000 parts loaded, 1,250 total runtime hours, and a contribution to production of more than 30,000 BMW X3 vehicles, with placement accuracy above 99% per shift and a cycle time of 84 seconds.

However, a Fortune investigation published in April 2025 reported that for much of the deployment only a single robot was performing one limited task, raising questions about how the broader characterization was supported. The BMW press release confirms the deployment involved two Figure 02 robots. Whether two robots running a single pick-and-place task for 11 months constitutes proof that humanoids can serve as a “flexible, reliable manufacturing workforce” depends heavily on how one defines “workforce.” The wording holds. The claim is doing more narrative work than the deployment data alone can sustain.

What the BMW deployment does not address, and what the claim does not mention, is the regulatory architecture that must be in place before that workforce can scale. That is where the analysis turns.

Three regulatory frameworks apply to Figure 03, and none fits a dynamically-stable humanoid without a bespoke bridge

Layers ordered top-down by jurisdictional urgency; ISO/WD 25785-1 is listed because its absence is the structural gap that makes the bridge approach necessary. EU AI Act obligations (separate from EU Machinery Regulation) and US state-level robotics regulations are named in the findings but not detailed here.

The Regulatory Terrain: Three Frameworks, Three Problems

ISO 10218:2025 and the Legged-Robot Gap

The revised ISO 10218:2025 industrial robot safety standard is the natural starting point for any manufacturer building a humanoid for factory use. There is, however, a problem with the standard itself: ISO 10218:2025 was written for statically stable industrial robots and explicitly excludes the mobility aspects of dynamically-stable legged and bipedal systems like Figure 03. The dedicated standard for legged humanoid robots, ISO/CD 25785-1, remains at committee-draft stage as of mid-2026 and has not been published; its scope covers robots that “could become unstable in the absence of power”.

The practical consequence is that any safety case built under ISO 10218:2025 alone is structurally incomplete for a dynamically-stable humanoid. Engineers must currently bridge this gap by overlaying ISO 12100 risk assessment methodology with an AI risk management framework, a bespoke approach that has not been validated by a published international standard and that different notified bodies may interpret differently.

OSHA Lockout/Tagout: The Balance Problem

OSHA’s lockout/tagout standard (29 CFR 1910.147) requires that equipment be de-energized before servicing. For a bipedal robot, this creates a hazard that the standard was not designed to contemplate: cutting power removes the active balance control that keeps the robot upright, meaning de-energization itself creates a new struck-by hazard from a falling robot. Standard LOTO procedures are therefore inapplicable in their traditional form.

The compliant alternative is OSHA’s alternative energy control procedure pathway, which requires a documented six-step controlled shutdown, mechanical bracing before battery isolation, specialized written procedures, and training not typical in standard factory safety programs. LOTO ranked fourth on OSHA’s FY2025 most-cited standards list, with 2026 penalties reaching up to $165,514 per willful violation. There is no public evidence that Figure AI has published or implemented a robot-specific LOTO procedure of this type for the BMW Spartanburg deployment or any other facility.

EU Machinery Regulation 2023/1230: The January 2027 Cliff

The pressure sharpens considerably for European expansion. The EU Machinery Regulation 2023/1230 becomes mandatory on January 20, 2027, replacing the Machinery Directive across all 27 member states with no national transposition variation and no grace period for new machine placements. For an AI-driven humanoid, it introduces five new Essential Health and Safety Requirements absent from the old Directive: protection of safety-critical software against cybersecurity attack; safety review of self-learning AI behavior (meaning a post-deployment Helix model update that changes robot behavior requires a new safety assessment); a mandatory human supervisory function; digital documentation with 10-year retention; and a provision that substantially modifying a machine, physically or via software update, can make the modifier the manufacturer with full CE marking obligations.

There is no public evidence that Figure AI has begun EU Machinery Regulation conformity assessment or CE marking preparation for Figure 03. BMW is reportedly evaluating expansion to European plants in Munich, Dingolfing, and Regensburg. If that expansion proceeds before a conformity assessment is complete, the legal exposure lands on the deployer as much as the manufacturer.

The Binding Constraint

Author analysis

The binding constraint is not production throughput, not the AI stack, and not capital. It is the absence of a published international safety standard governing the locomotion of dynamically-stable bipedal robots in human-collaborative environments, combined with documented evidence that Figure AI’s internal safety governance may not yet be adequate to build a credible case under the frameworks that do exist.

ISO/WD 25785-1 has not been published. Until it is, every humanoid manufacturer must construct a bespoke safety argument. That argument can be submitted to a notified body for CE marking, but its defensibility under enforcement scrutiny, and its portability across jurisdictions, is lower than an argument grounded in a ratified international standard. The regulation does not wait: January 20, 2027 is a hard date.

What would have to change for this constraint to stop binding? Two things, both independent of Figure AI’s engineering quality. First, ISO/WD 25785-1 must progress from working draft to published standard, a process that typically takes 18 to 36 months from working-draft stage and that Figure AI cannot accelerate unilaterally. Second, Figure AI must construct and have independently audited a conformity assessment that bridges the gap using ISO 12100 and an AI risk management framework, produce the technical file, and appoint a notified body for EU Machinery Regulation review. The January 2027 deadline is approximately six months away from the time of writing. The typical conformity assessment for a complex machine of this type takes longer than six months.

I have spent sixteen years designing systems whose safe state is de-energized. Cut power to a beam permit system and the shutters close, because springs close them and nothing has to work for that to happen. The whole discipline rests on the assumption that removing energy removes the hazard. A walking machine breaks that assumption, so the question is what replaces it.

If I were asked to sign off on this, I would require four things.

First, mechanical behavior on power loss that does not depend on software. Spring-applied, electrically-released brakes on every gravity-loaded joint, so the machine holds rather than collapses. That is ordinary practice on industrial arms. It is not sufficient here, because a braked biped mid-stride is a rigid body falling as one piece, which is worse than a compliant one. So brakes have to be paired with a defined mechanical descent path: rate-limited hip and knee release that folds the machine toward the ground rather than letting it topple, with the payload released before the descent begins unless releasing it makes things worse.

Xpeng's IRON, November 2025. Balancing actively with every step.Source: XPENG.

Second, an independent channel that stays alive when everything else dies. Its own energy reserve, its own processor, its own wiring, sharing no failure mode with the main battery or the main compute. Its only job is to reach the safe posture. Sized for a defined ride-through of a few seconds, and rated as a safety function with a performance level attached, not as a feature. In my world that layer is a separate cabinet with its own supply, and nothing in it is clever.

Third, a defined safe posture and, more importantly, a definition of where the machine is allowed to be when it cannot reach one. Kneeling or seated, lowest center of mass it can get to, impact taken on structure rather than on the head or the payload. Mid-step with 20 kg in its hands, it will not always get there. So the safety case cannot be a machine-level case. It has to be a zone-and-access case, the way a beam permit is: what is allowed in the space, who is in it, what interlocks the machine’s operating envelope when a person crosses a boundary. The machine alone cannot be made safe. The machine inside a defined access-controlled envelope can.

Fourth, servicing that never involves a technician approaching a free-standing biped. The service state is a fixture, not a floor. The machine walks into a cradle, the cradle latches, a positively-driven interlock confirms the latch, and only then does isolation happen. What the technician verifies is not zero energy, because zero energy is the hazard. The technician verifies mechanical restraint, then isolation, then stored energy dissipated, then a try-step. Under no circumstances is a machine serviced outside the fixture.

For the notified body, the argument I would build reaches past ISO 10218 entirely. IEC 61508 as the parent framework for the safety functions, ISO 12100 for the hazard analysis, ISO 13849 for the performance level of the descent channel, and then the closest ratified precedent for machines that must not fall on people, which is ISO 13482 for personal care robots, and the lift and hoist regimes where an uncontrolled descent has been a regulated hazard for a century. Evidence I would demand before signing: instrumented fall testing across every operating state including mid-step under load, contact and impact energies measured against the injury thresholds in ISO/TS 15066 and extended where 15066 does not reach, diagnostic coverage and failure rate data for the independent channel, and a pre-registered test matrix rather than a highlight reel.

Now the part I do not know how to solve. You cannot proof-test a fall. Everything I have certified depends on periodically exercising the safety function and watching it work: press the button, the shutter closes, sign the record. A controlled descent function cannot be exercised that way, because exercising it damages the machine. Test it rarely and your confidence in it decays between tests. Take credit for simulation instead and you have accepted a model as evidence for the one function that exists because models fail. I do not have a good answer to that, and I have not seen one published.

The Whistleblower Signal

The constraint is not merely theoretical. In November 2025, former Head of Product Safety Robert Gruendel filed a federal whistleblower lawsuit alleging wrongful termination after warning CEO Brett Adcock and chief engineer Kyle Edelberg that Figure 02 robots were powerful enough to fracture a human skull, citing internal impact testing showing forces more than double those required to break an adult skull. The complaint also alleged that Figure AI had “no formal safety procedures, incident-reporting systems, or risk-assessment processes” at the time of his hire, and that executives “gutted” a product safety roadmap prepared for two prospective Series C investors. A separate incident is described in which a robot malfunctioned and carved a deep gash into a stainless-steel refrigerator door.

Figure AI disputed the claims, stating Gruendel was terminated for poor performance. The lawsuit is unresolved, and its allegations are unproven. What the lawsuit establishes as a matter of public record, regardless of its outcome, is that the company’s most senior safety officer was terminated in September 2025, approximately five weeks before Figure 03 launched and two weeks before the Series C closed at $39 billion. There is no public evidence of who now holds the product safety function or whether the safety governance structure has been rebuilt.

Figure AI's safety governance structure changed materially in September 2025, and the current state has not been publicly disclosed

Safety Governance at Figure AI — Pre-September 2025

Robert Gruendel serving as Head of Product Safety (20+ year veteran)
Product safety roadmap prepared for Series C investors
Internal impact testing conducted on Figure 02
Incident-reporting system and risk-assessment process reportedly absent at hire (per Gruendel complaint)

Safety Governance at Figure AI — Post-September 2025

Gruendel terminated September 2, 2025
No public disclosure of replacement Head of Product Safety
No public safety certification (TÜV, UL, or equivalent) for Figure 03
No public conformity assessment under EU Machinery Regulation
80+ functional verification tests and 50+ inspection checkpoints per unit at BotQ (manufacturing quality system)

Derived from the Gruendel v. Figure AI complaint and public records; Figure AI disputes the allegations and states Gruendel was terminated for poor performance.

Technology and Production: What Is Working

Setting aside the regulatory gap, the engineering trajectory is real. Figure 03, launched October 9, 2025, stands 168 cm tall, weighs approximately 60 to 61 kg, carries 20 kg payloads, and operates on a 2.3 kWh battery for four to five hours. Its 40 degrees of freedom, palm-embedded cameras, and tactile fingertip sensors capable of detecting forces as small as 3 grams represent a meaningful hardware step from its predecessor.

Two Figure 03 humanoid robots in a warehouse aisle. The nearer one, in a grey textile suit with a black reflective visor, holds a large, dark automotive trim panel in both hands. Green plastic bins line the shelves, and purple light rails run along the floor. — Figure 03 sequencing parts at the BMW Plant Spartanburg. The deployment behind the claim.Source: BMW Group.

The AI architecture has evolved in parallel. Helix, Figure’s proprietary vision-language-action model, was publicly introduced in February 2025 after the company ended its OpenAI model partnership. Helix 02, released in January 2026, added a third processing layer executing at 1 kHz to handle balance, contact forces, and whole-body coordination, trained on over 1,000 hours of human motion data and replacing hand-engineered locomotion controllers.

Manufacturing has scaled faster than most observers expected. BotQ, Figure’s San Jose facility, reached one Figure 03 per hour by late April 2026, up from one per day in January, a 24-fold throughput improvement in under 120 days. Over 350 units had been delivered by that point, with end-of-line first-pass yield above 80% and battery line yield at 99.3% across 500-plus packs. Each robot passes 50-plus in-process inspection checkpoints and 80-plus functional verification tests before shipping. That quality system is an early building block for a safety case, but it is a manufacturing quality system, not a conformity assessment against a safety standard.

From Series A onward each round was larger than the last, and the Series C at $1B+ exceeded all prior rounds combined

Round sizes in USD millions; Series B had no publicly named lead investor per the official Figure AI press release.

Financial Position and the Valuation Question

Figure AI has raised approximately $1.9 billion, including $675 million in its Series B in February 2024 and more than $1 billion in a Series C that closed September 16, 2025 at a $39 billion post-money valuation, with PitchBook putting the total at $2.34 billion and Tracxn at $1.75 billion on differing treatments of founder seed capital and secondary activity. Whatever is constraining this company, it is not capital.

Cumulative capital raised crossed $1.8 billion after the Series C, with most of it raised since early 2024

Cumulative totals by round; figures per Sacra/Yahoo Finance estimate; PitchBook reports $2.34B and Tracxn $1.75B depending on treatment of founder seed capital and secondary transactions.

Competitive Context

Agility’s Digit has the strongest commercial deployment record, with 65,000-plus operating hours across nine facilities. Chinese manufacturers, principally AgiBot and Unitree, accounted for roughly 90% of the estimated 16,000 to 18,000 global humanoid units shipped in 2025, with Unitree pricing starting at $4,900. The cost differential is not a narrow one, and it is the vector most likely to pressure Figure’s industrial pricing over time.

The auto-industry partnership pattern is instructive. Every major automaker except Tesla is partnering with a specialist humanoid robotics firm: BMW with Figure AI, Hyundai with Boston Dynamics, Mercedes-Benz with Apptronik, Toyota with Agility Robotics. That structure concentrates both the commercial opportunity and the reputational risk. A safety incident at BMW Spartanburg would land on both companies simultaneously.

What Would Change This

Author analysis

Arguments that would increase confidence that the BMW claim holds at scale are: publication of ISO/WD 25785-1, which would give Figure AI and its customers a ratified standard to certify against; public disclosure of a completed EU Machinery Regulation conformity assessment and CE marking for Figure 03; independent third-party safety audit (TÜV, UL, or equivalent) of Figure 03’s human-collaborative operating mode; and resolution of the Gruendel lawsuit in a manner that discloses Figure AI’s current safety governance structure.

Arguments that would reduce confidence: a safety incident at BMW Spartanburg or any subsequent deployment resulting in OSHA recordable injury; enforcement action under the EU Machinery Regulation against a European Figure AI deployment; the Gruendel lawsuit producing discovery disclosures that corroborate the alleged absence of formal safety procedures; or BMW choosing a different vendor for European expansion, which would signal that the safety certification gap was a real barrier rather than a theoretical one. The Fortune reporting that for much of the Spartanburg deployment only one robot was performing one limited task already points in this direction.

The most important asymmetry is that the regulatory clock is not synchronized with the engineering clock. Figure AI can ship 240 robots a month regardless of where ISO/WD 25785-1 stands. The EU Machinery Regulation does not share that flexibility.

Author analysis

The Annex I Question

Two facts reported separately in this piece connect in a way that, as far as the public record shows, nobody has connected. Figure’s own description of Helix 02 states that a third processing layer executing at 1 kHz now handles balance, contact forces, and whole-body coordination, and that it replaced hand-engineered locomotion controllers. The EU Machinery Regulation’s Annex I Part A lists, in item 5, safety components with fully or partially self-evolving behaviour using machine learning approaches ensuring safety functions; item 6 covers machinery embedding them.

For a dynamically-stable biped, balance is not a performance feature. Falling is the hazard, and the balance controller is the function that prevents it. If the learned layer is the component that keeps the machine upright, then on a plain reading Figure 03 already contains an Annex I Part A component. That routes the machine to mandatory third-party conformity assessment rather than self-declaration, and it makes every Helix update that changes balance behaviour a modification to a safety-related component, under a regulation that treats substantial modification as re-manufacture.

The reading has an escape. Figure has not disclosed how its safety functions are partitioned. If a deterministic supervisor owns the stability claim, with the learned layer advising inside envelopes the supervisor enforces, the balance controller is not self-evolving in the regulation’s sense and the reading fails. That partition is precisely what a conformity assessment would have to establish, and it is not public.

The verdict

The Verdict

Verdict: Undetermined

Confidence: Low, for reasons of evidence rather than judgment. The evidence base on Figure AI’s safety governance is thin, single-sourced in critical places (primarily the Gruendel complaint and the absence of any public certification record), and the company has not disclosed the information that would settle the question. The production data is solid. The engineering trajectory is credible. The regulatory gap is documented and independent of Figure AI’s own disclosures.

The BMW claim, taken on its operational terms, is broadly supported by the deployment data, with the caveat that a two-robot, single-task pilot over 11 months is a narrow empirical base for the word “proved.” The broader question this series addresses, whether the safety case can be made to a defensible standard, cannot be answered affirmatively on current evidence.

The disclosures that would settle it are: (1) a published ISO/WD 25785-1 standard with a ballot date, which would surface through ISO’s public standards tracking portal; (2) a CE marking declaration of conformity for Figure 03 under EU Machinery Regulation 2023/1230, which becomes a public document upon placement on the EU market; (3) an OSHA inspection record from Plant Spartanburg, accessible via OSHA’s enforcement database, that either cites or clears Figure AI’s energy control procedures; and (4) the outcome of Gruendel v. Figure AI (No. 5:25-cv-10094, N.D. Cal.), whose docket is public through PACER.

Observables:

First, ISO/WD 25785-1 ballot status. The ISO standards tracking portal will show whether this document has advanced from working draft to committee draft or draft international standard by Q2 2027. If it has not, the compliance gap for all humanoid manufacturers remains open, and any safety case built before that date remains structurally provisional.

Second, EU Machinery Regulation CE declarations for Figure 03. If BMW proceeds with European deployment in Munich, Dingolfing, or Regensburg before January 20, 2027, the CE marking and conformity assessment would need to be in place, and the declaration of conformity is a public document. If no such document appears in EU market surveillance databases by Q1 2027, European expansion has either been delayed or is proceeding without full compliance.

Third, OSHA enforcement records for Plant Spartanburg. OSHA’s inspection history for BMW’s South Carolina facility is publicly accessible. Any inspection triggered by a robotics-related incident or a whistleblower referral would appear there, typically within 90 to 180 days of the inspection’s close.

Fourth, Gruendel v. Figure AI docket activity. Discovery in this case, if it proceeds, would compel disclosure of Figure AI’s internal safety procedures, incident logs, and risk assessment records. Docket entries are public through PACER and typically surface within days of filing.

Fifth, the safety-function partition for Helix 02. Any public description of Figure 03’s safety architecture, whether in CE technical documentation, a notified body engagement, a published safety white paper, or discovery in Gruendel v. Figure AI, that establishes whether balance is owned by a deterministic layer or a learned one. If the learned layer owns it, Annex I Part A applies, and the conformity route changes. Figure can settle this with a single disclosure at any time.

This verdict is on the ledger

Sources (51)

BMWGROUP
BMW Group advances use of Physical AI with Figure 03 at Spartanburgpress.bmwgroup.com
FIGURE
Figure Exceeds $1B in Series C Funding at $39B Post-Money Valuation — Figure AI Officialfigure.ai
PRNEWSWIRE
Figure Exceeds $1B in Series C Funding at $39B Post-Money Valuation — PR Newswireprnewswire.com
INTELCAPITAL
Figure Exceeds $1B in Series C Funding at $39B Post-Money Valuation — Intel Capitalintelcapital.com
PRNEWSWIRE
Figure Raises $675M at $2.6B Valuation and Signs Collaboration Agreement with OpenAI — PR Newswireprnewswire.com
SACRA
Figure AI valuation, funding & news | Sacrasacra.com
WIKIPEDIA
Figure AI — Wikipediaen.wikipedia.org
FORGEGLOBAL
Forge Global secondary-market price page — Figure AIforgeglobal.com
FIGURE
Introducing Figure 03 — Figure AIfigure.ai
FIGURE
Helix: A Vision-Language-Action Model for Generalist Humanoid Control — Figure AIfigure.ai
FIGURE
Introducing Helix 02: Full-Body Autonomy — Figure AIfigure.ai
FIGURE
Scaling Helix: a New State of the Art in Humanoid Logistics — Figure AIfigure.ai
THEROBOTREPORT
Figure humanoid robots use Helix VLA model to demonstrate household chores — The Robot Reporttherobotreport.com
INTERESTINGENGINEERING
Figure robot gets AI brain that enables human-like full-body control — Interesting Engineeringinterestingengineering.com
FIGURE
BotQ: A High-Volume Manufacturing Facility for Humanoid Robots — Figure AIfigure.ai
FIGURE
Ramping Figure 03 Production — Figure AIfigure.ai
THEROBOTREPORT
Figure AI unveils BotQ high-volume humanoid manufacturing facility — The Robot Reporttherobotreport.com
ROBOTICSANDAUTOMATIONNEWS
Figure ramps up humanoid robot manufacturing at unprecedented speed — Robotics & Automation Newsroboticsandautomationnews.com
FIGURE
F.03 Arrives at BMW — Figure AIfigure.ai
REPAIRERDRIVENNEWS
Humanoid robots complete 11-month project at BMW plant — Repairer Driven Newsrepairerdrivennews.com
IIOT-WORLD
Physical AI Deployment ROI: BMW's 30,000-Car Proof — IIoT Worldiiot-world.com
THEROBOTREPORT
BMW Group deploys Figure 03 humanoid after tests with previous version — The Robot Reporttherobotreport.com
ISO
ISO 10218-1:2025, Robotics — Safety requirements — Part 1 (ISO catalogue entry)iso.org
ISO
ISO/CD 25785-1, draft under committee review (ISO catalogue entry)iso.org
THEROBOTREPORT
ISO 10218 industrial robot safety standard receives major overhaul — The Robot Reporttherobotreport.com
OSHA
29 CFR 1910.147, The control of hazardous energy (lockout/tagout) — OSHAosha.gov
EUROPA
Regulation (EU) 2023/1230 on machinery — EUR-Lexeur-lex.europa.eu
EFS
EU Machinery Regulation 2023/1230 — Requirements, obligations and implementation — EFS Consultingefs.consulting
CNBC
Figure AI sued by whistleblower who warned that startup's robots could 'fracture a human skull' — CNBCcnbc.com
COURTLISTENER
Gruendel v. Figure AI, Inc., No. 5:25-cv-10094 (N.D. Cal., filed November 21, 2025) — CourtListener docketcourtlistener.com
INTERESTINGENGINEERING
Figure robot hand could 'fracture a human skull,' whistleblower warns — Interesting Engineeringinterestingengineering.com
ROBOTICSANDAUTOMATIONNEWS
Former Figure AI engineer claims company's humanoid robots 'powerful enough to fracture human skull' — Robotics & Automation Newsroboticsandautomationnews.com
DDG
Grendel v. Figure AI: whistleblower vs. robot safety risks — DDGddg.fr
WIKIPEDIA
Brett Adcock — Wikipediaen.wikipedia.org
FIGURE
Master Plan — Figure AIfigure.ai
IEEE
Humanoid Robots: Robotics Experts Found Persona AI Startup — IEEE Spectrumspectrum.ieee.org
TECHCRUNCH
Persona's founders are certain the world can use another humanoid robot — TechCrunchtechcrunch.com
CONTRARY
Report: Figure Business Breakdown & Founding Story — Contrary Researchresearch.contrary.com
SOURCERY
BREAKING: Brett Adcock, CEO of Figure — Sourcerysourcery.vc
DLAPIPER
DLA Piper advises Parkway Venture in its participation in US$675 million Series B investment in Figure AIdlapiper.com
MARKETSANDMARKETS
Humanoid Robot Market Size, Share, Latest Trends & Growth Analysis, 2026–2035 — MarketsandMarketsmarketsandmarkets.com
GRANDVIEWRESEARCH
Humanoid Robot Market Size, Share & Growth Report, 2033 — Grand View Researchgrandviewresearch.com
FORTUNE
Is Figure AI's BMW robot partnership what its founder Brett Adcock says? — Fortunefortune.com
TRACXN
Figure — 2026 Company Profile, Team, Funding & Competitors — Tracxntracxn.com
ROBOTICSANDAUTOMATIONNEWS
Inside Figure AI: How this startup is reshaping the future of humanoid robots — Robotics & Automation Newsroboticsandautomationnews.com
ROBOTICS247
General-purpose humanoid developer Figure AI exceeds $1 billion after Series C — Robotics 24/7robotics247.com
TECHCRUNCH
Figure reaches $39B valuation in latest funding round — TechCrunchtechcrunch.com
INTERESTINGENGINEERING
US: BMW deploys Figure 03 humanoid at smart factory for logistics — Interesting Engineeringinterestingengineering.com
REVELIOLABS
Figure AI Number of Employees 2026 — Revelio Labsreveliolabs.com
COUNTERPOINTRESEARCH
Figure AI Unveils its 2nd-Gen Robot — Counterpoint Researchcounterpointresearch.com
FORBES
Humanoid Robots in Manufacturing: Timelines, Cost, And Opportunity — Forbesforbes.com