Tutorials

Your Car Is a Sensor Network. Canada Is Finally Asking Who Controls It.

Canada's Chinese EV review exposes a bigger security question: who controls a connected car's sensors, cloud data, software updates, and remote access?

Approximately 7 min read

A car can map its surroundings, record its location, pair with a phone, receive software updates, and communicate with its manufacturer’s cloud. Some of those capabilities are useful to the driver. In the wrong trust relationship, the same capabilities create an intelligence-gathering surface.

On November 18, 2026, Global News reported that Canada’s transport minister, Steve MacKinnon, had put potential limits on Chinese EV access to sensitive military and government sites within a broader cross-government review. This was a review, not an announced nationwide driving ban. The distinction matters: no published result in that reporting establishes that every Chinese EV is exfiltrating sensitive information.

The timing is unusual. Canada implemented a 49,000-vehicle annual quota at a 6.1% tariff for qualifying Chinese EV imports on March 1, 2026. The government is opening a trade channel while asking how to defend sensitive environments from the connected systems travelling through it.

The sharper question is not whether a vehicle was assembled in China. It is who can access its data, change its code, or issue commands to it after the vehicle has been sold.

The security perimeter is not the parking gate

Preventing an untrusted vehicle from entering a secure compound is a reasonable physical-security measure. It can reduce close-range observation, access to restricted routes, and the exposure of sensitive personnel or facilities.

But a gate does not control a vehicle’s cloud account, OTA update infrastructure, telematics service, or data retention. Nor does it prevent a vehicle travelling on a public road outside the site from collecting ordinary sensor or location data. What information is actually gathered or uploaded depends on the vehicle’s model, configuration, services, and operational behaviour. A camera’s existence does not prove continuous recording; GPS capability does not prove a foreign government receives a location feed.

This is why a location restriction may be a useful layer without being a complete cybersecurity control.

Canada’s own vehicle cybersecurity strategy already describes modern vehicles as connected cyber-physical systems, with cellular, Bluetooth, Wi-Fi, onboard diagnostics, infotainment, and telematics interfaces. Its scope is technology-neutral. That is a stronger starting point for a security assessment than a blanket assumption about one manufacturer’s logo.

Separate the three systems that are often called ‘the car’

A meaningful review should distinguish three layers. They have different attack paths and require different evidence.

Layer What is at risk? What would establish control?
Sensors and onboard storage Camera images, vehicle location, route history, cabin and device information Inspect actual data collection, retention and local access settings
Connectivity and cloud services Telemetry uploads, paired-phone information, account-linked tracking, third-party sharing Map destinations and service operators; examine payload types, permissions and backend access
Software and remote operations OTA updates, remote diagnostics, remote vehicle functions, changing code after sale Review update signatures, signing-key custody, role permissions, audit records and rollback protections

These are questions for an assessment, not findings about a specific automaker. Encrypted traffic can hide the payload from a network observer, and seeing an endpoint does not establish that it receives camera footage. Conversely, a clean packet capture during a short drive is not proof that a dormant or event-triggered upload path does not exist.

The control plane deserves special attention. A car may behave acceptably today while still relying on a supplier-controlled update service that can change software tomorrow. If the same remote organization can deploy code, extract diagnostics and administer accounts, the trust boundary extends far beyond the dashboard.

Why jurisdiction matters without proving espionage

Country of assembly, supplier origin, legal jurisdiction and actual operational control are different properties. A Western-branded vehicle can contain foreign-developed connectivity components. A Chinese-branded vehicle can use cloud and service arrangements that vary by market. Neither the badge nor the factory location is a complete software bill of materials.

Still, jurisdiction is not irrelevant. Governments care whether a manufacturer or critical supplier can be compelled to provide data, alter systems, or enable remote access. The U.S. Bureau of Industry and Security connected-vehicle rule explicitly addresses certain software, vehicle connectivity hardware and manufacturers with sufficient links to China or Russia. It applies phased restrictions, including to covered software and manufacturers for model year 2027 and covered connectivity hardware for model year 2030.

That is a supply-chain regulatory approach, not evidence that a particular vehicle has conducted spying. It also should not be confused with a universal prohibition on a Canadian driver crossing the U.S. border in any Chinese-built car. Import, sale and temporary entry are different legal questions.

There is a real policy tension here: a jurisdiction-based restriction can address risks that a packet capture will never reveal, such as compelled future cooperation. A purely technical test can expose unsafe implementation regardless of the country of origin. A credible program needs both lenses, without pretending they are interchangeable.

Five questions Canada could require manufacturers to answer

For civilian buyers and public-sector fleet operators, the most useful outcome would be a reproducible, model-specific security disclosure—not a vague assurance that vehicles are safe.

  1. What leaves the vehicle? Publish the categories of telemetry, imagery, diagnostics and paired-device data that can be transmitted, including defaults, triggers, retention periods and deletion mechanisms.
  2. Who can retrieve it? Identify cloud operators, third-party processors, account administrators and legal jurisdictions. Distinguish storage location from the entities with administrative or cryptographic access.
  3. Who can change the software? Document OTA signing, key custody, release approval, emergency update powers, rollback protections and the boundary between infotainment and safety-critical systems.
  4. What works offline? Describe which functions remain usable when connectivity or cloud services are disabled, what telemetry is buffered locally, and whether queued data is uploaded later.
  5. What can be independently checked? Provide an assessment path for firmware inventory, network behaviour, role-based access controls, vulnerability disclosure and security updates. A one-time certification is not enough for a product whose software changes throughout its life.

A useful evidence package would record the exact model, market version, firmware build, app version, account region, network conditions and test date. Compare repeatable tests under matched settings; otherwise two reviewers may be evaluating different cars in everything but their model name.

These proposed requirements are not claims that Canadian regulators have adopted them. They are an engineering framework for translating the policy debate into observable controls.

Three different decisions, not one national yes-or-no

For a private commuter buying an EV, the relevant controls may be transparent data settings, secure updates, limited mobile-app permissions and a clear support lifecycle.

For an enterprise fleet, the review should add account administration, fleet telematics exports, third-party processors, device pairing, incident response and the ability to remove employee data. The Canadian Centre for Cyber Security already recommends fleet vulnerability assessments, security updates, access control and separation from sensitive organizational networks.

For a military installation or other high-sensitivity site, local access policies may need to be stricter. Those rules can consider whether any connected vehicle—regardless of brand—should enter specified areas with active cameras, connectivity or diagnostic capabilities. A risk tier makes more sense than pretending every road and every parking lot handles the same information.

There is no universal switch that makes a modern connected car non-networked, and disabling networking can also affect emergency or safety-related services. Restrictions and test procedures must account for those trade-offs rather than improvising operational changes on public roads.

What evidence would actually change the conclusion?

A credible security finding would look like a repeatable capture of sensitive data leaving the vehicle without appropriate authorization; documented backend access that exceeds the stated permissions; an independently demonstrated failure of update authentication; or a validated remote-control weakness.

A marketing brochure is not that evidence. Neither is the presence of Chinese-made hardware, the existence of a cellular modem, or a list of foreign IP addresses without traffic context.

Canada has a chance to publish a more useful standard of proof. Its own parliamentary briefing materials already recognize that connected vehicles collect data with potential intelligence value. The next step is to define what manufacturers must disclose, what independent reviewers can test, and which residual risks justify restricting sensitive-site access.

The automotive security question of 2026 is not simply who built the car. It is who controls the software-defined vehicle after it leaves the showroom.


Evidence basis: November 18 Global News reporting; Canadian import policy and government cybersecurity guidance; and the U.S. BIS connected-vehicle rule. This is a policy and architecture analysis, not an independent test of BYD, Tesla, or any other manufacturer’s vehicle.

Sources and further reading

Continue reading

More in Tutorials ↗