Profile of Tencent's Keen Lab which exposed a number of flaws in Tesla Autopilot over the years, most recently by using stickers on lane markings
Iris Deng / South China Morning Post :
Context & Ripple Effects
Days after researchers showed that stickers on lane markings could steer Tesla Autopilot into oncoming traffic — a flaw Tesla says it has patched — the South China Morning Post profiles the team behind it: Keen Lab, Tencent's security research unit with a multi-year record of exposing Autopilot weaknesses. The profile lands at an awkward moment for Tesla, whose self-driving claims were already drawing scrutiny.
The deeper wrinkle is Tencent's dual position: Bloomberg reported in 2017 that Tencent was developing its own autonomous driving technology and testing a prototype internally, meaning the company running the most prominent adversarial research on Tesla's driver-assistance stack is also building a competing one.
First-order effects
- Tesla has shipped a patch for the sticker attack, but the profile keeps attention on the fact that vision-only lane-keeping was fooled by physical-world tampering — a direct reputational hit to Autopilot's reliability claims.
- Keen Lab cements its standing as the field's most consistent external red-teamer of production driver-assistance systems, with each disclosure raising the bar for what Tesla must defend against.
Second-order effects
- Tencent's own autonomy program benefits from the same research: techniques proven against Autopilot inform both its defensive engineering and its competitive read on rival systems, blurring the line between security disclosure and competitive intelligence.
- Every automaker shipping assisted-driving features now faces pressure to commission comparable adversarial testing before regulators or researchers do it for them, making third-party security audits a de facto requirement.
Third-order effects
- The pattern that follows in this corpus — lawsuits claiming Tesla dangerously overhyped its self-driving software, then a whistleblower leak of 100GB of customer FSD complaint data, and later reporting that crash-data analysis shows Tesla exaggerating FSD's safety via flawed methodology — points toward a regime where vendor safety claims are systematically contested by outside evidence rather than accepted on assurance.
- If independent red-teaming and leaked internal data keep outrunning official claims, driver-assistance marketing will migrate toward verifiable, externally audited benchmarks as the basis for regulatory approval and consumer trust.
The trend: Autonomous-driving safety is shifting from vendor-declared capability to contested evidence, with adversarial labs, litigants, and whistleblowers supplying the counterweight to manufacturer claims.