Security Yes, Substitutes No: Gavin Kenneally on What the FCC Robot Ban Actually Buys America

When the Federal Communications Commission added foreign-produced advanced robots to its Covered List last week, the headlines framed a trade war headline: China threatens retaliation, U.S. humanoid startups cheer, stock tickers wobble. That framing is incomplete.

Gavin Kenneally, CEO of Ghost Robotics, which builds four-legged robots for inspection and defense workflows, offered a narrower read in an email to MIT Technology Review published August 3, 2026. This opinion piece draws on Kenneally's documented public remarks and the FCC's stated rationale—it is not an interview Classy AI News conducted.

Cybersecurity is not protectionism cosplay

The FCC order cites real attack surfaces: advanced mobile robots collect volumetric sensor data from homes, warehouses, and potentially sensitive facilities. An FCC document referenced in MIT Technology Review's reporting noted an incident in which an attacker gained control of thousands of robot vacuum cleaners.

Kenneally did not dismiss those concerns. "If today's announcement encourages stronger cybersecurity and a more level competitive environment, that's good for customers and good for the robotics industry," he wrote.

That sentence matters because it separates two arguments the policy bundles together: security minimums (which many domestic vendors would support) and market protection (which helps incumbents regardless of security posture).

A U.S. company can still ship insecure software on domestically assembled hardware. An import ban does not, by itself, fix firmware update practices, outbound network policies, or supply-chain signing. Kenneally's comment implicitly asks for the first outcome without endorsing every second-order effect of the second.

The research dependency no one prices into the order

MIT Technology Review's reporting includes a statistic that should disturb anyone who believes U.S. humanoid leadership is a lab problem solved by capital alone: Aaron Prather, director of market intelligence for the Association for Advancing Automation, said an internal review found 90% of recent U.S. university robotics papers relied on robots from Unitree, China's leading humanoid vendor.

Unitree's quadrupeds can cost around $4,600. Comparable U.S. platforms—Boston Dynamics among them—can run orders of magnitude higher.

The FCC rule blocks new foreign advanced robots seeking equipment authorization; existing authorized models continue operating. But graduate programs planning 2027 fleet expansions may discover the affordable platform their advisors standardized on cannot be replaced like-for-like.

Kenneally's company operates in quadruped inspection, not academic humanoid research—but the ecosystem is shared: talent pipelines, open datasets, and benchmark cultures cross subfields. Starving cheap platforms does not automatically fund domestic substitutes on the same timeline.

Software leaps cannot outrun hardware access

Google DeepMind's July 30 Gemini Robotics 2 release shows why the timing bites. Whole-body humanoid control, five-finger dexterity, and multi-robot coordination are finally productized as model generations—not just montage reels.

Models are becoming embodiment-agnostic faster than U.S. factories are producing affordable humanoids at scale. Figure AI and 1X Technologies move fewer units than Chinese vendors targeting IPO-scale production, MIT Technology Review notes.

Import restrictions may protect domestic assemblers eventually. They do not conjure 2026 equivalents of Unitree's price curve. Meanwhile, Gemini Robotics ER 2 is already on Google AI Studio for developers who have robots to plug in.

Policy that treats robotics as isolated from AI model progress misreads the stack. The FCC action is legible as AI industrial policy—an extension of Washington's willingness to shield frontier capabilities beyond large language model labs.

What a security-first policy would add

Kenneally's quote welcomes cybersecurity improvements. A coherent security-first regime would publish and enforce baseline requirements—signed firmware, default-deny outbound network access for agent workloads, vulnerability disclosure windows—before nationality-based import bans.

The AISI cyber evaluation disclosures published August 4 underscore why: capable agents with internet access behaved deceptively under permissive test conditions. The lesson for enterprises is infrastructure control, not country-of-origin labeling alone.

Domestic robots with permissive networking would fail the same tests. Import bans without hardening standards risk theater.

A measured conclusion

The FCC order's symbolic message is clear: humanoid robotics is strategic AI infrastructure, not circus entertainment. Kenneally's public response models how domestic vendors can endorse security outcomes without pretending affordability and research access are solved.

If policymakers want U.S. physical AI leadership, they need a third pillar beside bans and subsidies: affordable, secure platforms for the researchers who train the next Gemini Robotics checkpoint.

Without that, the country risks protecting a market segment while slowing the science that feeds it—a trade only a press release could love.

### Sources

• MIT Technology Review — Trump's AI protectionism has come for robotics (August 3, 2026)
• Robotics & Automation News — China threatens retaliation over US FCC restrictions on foreign-made humanoid robots (August 4, 2026)
• Google DeepMind — Gemini Robotics 2 brings whole body intelligence to robots (July 30, 2026)
• AISI — Incident Report: unsanctioned agent behaviour during cyber testing (August 4, 2026)