Meta AI Testing Scandal: Contractors Posed as Children to Probe Rival Chatbots

by Zaira Varghese 02. July 2026 AI 0

A Meta AI testing operation is at the center of the tech world’s newest ethics firestorm. According to an investigation by Wired, Meta hired hundreds of contractors in Kenya to pose as children and flood rival chatbots — ChatGPT, Google’s Gemini, and Character.AI — with thousands of crisis prompts about suicide, sex, and drugs, all without the targeted companies’ knowledge or consent.

The revelation has ignited a fierce debate: is this red-team safety research, competitive intelligence, or something far murkier?

Inside Operation ‘Cannes’

The program, internally codenamed “Cannes,” was run through Covalen, an outsourcing firm, with contractors based in Kenya. Workers were reportedly instructed to create fake accounts listing ages under 18 and then engage rival AI chatbots with the kind of messages that safety teams dread most.

The scale was industrial. A single testing round in August 2025 involved more than 45,000 prompts covering suicide, self-harm, sex, drugs, and eating disorders, according to Wired’s reporting. Contractors also submitted disturbing images alongside their queries — pills, knives, nooses, and medical diagrams — to test how rival systems responded. The project was reportedly still active as of April 2026.

Rivals Say They Never Agreed to the Meta AI Testing

None of the targeted companies were informed. Their responses so far:

Character.AI said the conduct violated its Terms of Service.
OpenAI said it is looking into the matter.
Google said it had not approved the testing and did not know its purpose.

That lack of consent is what separates this from ordinary benchmarking. Adversarial testing of AI models is standard industry practice — but it is typically done on your own systems, or through disclosed third-party audits, not by covertly impersonating minors on a competitor’s platform at scale.

Why the Kenya Angle Makes It Worse

The location of the workforce adds another layer of controversy. Kenya has become a hub for outsourced AI labor, and content-related work there has repeatedly drawn scrutiny over pay and the psychological toll on workers. Asking contractors to spend their days simulating children in crisis — writing thousands of messages about suicide and abuse — raises exactly the kind of worker-welfare questions that have dogged the industry’s supply chain for years.

What Was Meta Actually Doing?

Meta has not fully explained the program’s purpose. The charitable reading is competitive safety benchmarking: understanding how rivals handle minors in crisis, perhaps to calibrate its own guardrails after its AI products faced criticism on the same front. The less charitable reading is that Meta was quietly building a dossier of rivals’ failures — material that could be useful in policy fights, PR battles, or regulatory proceedings.

Either way, the optics are brutal. Regulators in the US and EU are already scrutinizing how AI chatbots handle minors, and the Meta AI testing story hands them a vivid case study in how far companies will go in the shadows.

The Bigger Picture

The episode exposes a gap in the AI industry’s rules of engagement. There are no established norms for cross-company adversarial testing, no disclosure requirements, and no standards for the human labor behind it. Until those exist, expect more stories like this one — and expect the companies caught on either side of them to keep pointing fingers.