Meta pays contractors to pose as teens, probe rival chatbots
WIRED reports Meta ran a contractor program called "Cannes" that paid people to pose as teens and feed rival chatbots thousands of sexual, suicidal and disturbing prompts. Meta calls it safety testing; critics call the scale and method alarming.
Meta paid hundreds of contractors to pose as under‑18 users and bombard competing chatbots with sexual, suicidal and other disturbing prompts in a covert testing program, according to a WIRED investigation published on June 8, 2026. The project, codenamed “Cannes,” ran through a contractor called Covalen and — in at least one round — generated more than 45,000 test prompts, WIRED reports.[2]
The reporting raises questions about how far platforms should go when stress‑testing competitors' systems and whether contractor‑led campaigns that mimic minors cross ethical or legal lines. Meta says the work was safety testing and industry‑standard; critics described the scale and content as alarming.
Project 'Cannes' and the scale of prompts
WIRED reviewed a spreadsheet tied to the program that contained 3,748 individual prompts and described broader rounds exceeding 45,000 prompts aimed at multiple rival systems, including OpenAI’s ChatGPT, Google’s Gemini, and Character.AI.[2] The prompts reportedly asked chatbots for instructions about suicide and self‑harm, details about sex and eating disorders, and referenced graphic images such as nooses, knives and pills.[2][1]
Futurism summarized the reporting as alleging “hundreds of contractors” were paid to pose as teenagers and escalate content to test responses across platforms.[1] Social media threads amplifying the WIRED story included screenshots and contractor accounts posted to Reddit and Instagram, though those posts are user‑generated and not independently verified.[6][3][11]
Meta's defence and the 'industry‑standard' claim
Meta issued a statement to WIRED saying, “Testing and benchmarking chatbot responses to help ensure safe and age‑appropriate experiences is a responsible, industry‑standard practice, and any suggestion otherwise completely misunderstands how technology companies work to refine and improve their systems.”[2][5] The Times of India also carried Meta’s response, which framed the activity as part of normal safety work.[5]
Meta declined to provide additional documentation to WIRED for independent verification of the complete dataset, the reporting says.[2] That gap leaves unanswered questions about internal oversight, the instructions given to contractors, and whether any safeguards were in place to limit exposure to extreme content.
Which rivals were targeted and why now
WIRED names OpenAI, Google and Character.AI among those whose chatbots were tested, and the spreadsheet and contractor materials suggest the testing focused on eliciting age‑specific and high‑risk responses.[2] The timing — appearing amid intensifying competition over assistant quality and safety — suggests Meta was seeking comparative data to inform its own AI safety and moderation work. But the method also risked creating disturbing logs and placing contractors in contact with traumatic material.
The reporting prompted debate on social platforms, where some commenters called the tests a legitimate stress‑test, while others said pretending to be minors and pushing graphic prompts was ethically dubious. Reddit threads flagged by users amplified concerns about contractor welfare and the potential for such datasets to be misused.[6]
Critics contacted in the wake of WIRED’s piece described the approach as blunt and risky; however, major outlets such as Reuters, Bloomberg and the Financial Times have not, as of publication, independently corroborated WIRED’s full findings. That lack of additional Tier‑1 confirmation means the claims should be treated as reported by WIRED and echoed by other outlets rather than conclusively established.[2]
Meta’s previous contractor‑run data programs have drawn scrutiny before; the company has long relied on third‑party vendors for content labelling and safety testing. Past controversies over contractor exposure to harmful content and transparency lapses suggest this episode may revive calls for clearer disclosure and stronger protections for human reviewers.
Closing paragraph: what to watch
Expect follow‑up scrutiny from journalists and regulators: the next concrete items to watch are whether Reuters or the FT corroborate WIRED’s dataset, whether Meta releases internal audit materials about the Cannes program, and whether contractors or oversight bodies lodge formal complaints that trigger regulatory or legal action.
