Meta Contractors Posed as Teens to Test Rival AI Chatbots; US-Iran Tensions Flare Over Hormuz

The AI safety debate took a sharp turn today after WIRED reported that Meta paid hundreds of contractors to pose as teenagers and stress-test rival chatbots from OpenAI, Google, and Character.AI ^[1]. The project, internally named Cannes and managed by Meta contractor Covalen, created dummy under-18 accounts to send prompts about suicide, sex, drugs, eating disorders, and self-harm to competitors' systems ^[2]. One round in August 2025 alone pushed more than 45,000 prompts through ChatGPT, Gemini, and Character.AI without those companies' knowledge ^[2]. A reviewed spreadsheet of 3,748 prompts included hundreds on suicide and self-harm, hundreds more on eating disorders, and at least 239 on sex or romance, with some written from the voice of a child in crisis ^[2]. Meta has defended the effort as routine safety work ^[1].

The operation raises fresh questions about how AI giants police each other and whether covert red-teaming crosses into corporate espionage. While red-teaming is a standard safety practice, doing so through fake minor accounts and without disclosure to competitors blurs the line between safety research and surveillance ^[1]. The report lands as regulators in the U.S. and Europe are already scrutinizing child safety and content moderation across generative AI platforms.

Meanwhile, world news is dominated by fragile diplomacy in the Middle East. U.S. envoys Steve Witkoff and Jared Kushner arrived in Doha, Qatar, on Tuesday for indirect talks about the interim U.S.-Iran deal brokered by Pakistan and Qatar ^[3]. Qatar's Foreign Ministry spokesman Majed al-Ansari said the envoys would meet mediators, not Iranian officials, and that no high-level direct talks were scheduled ^[4]. The visit follows a weekend of crossfire in the Persian Gulf over reopening the Strait of Hormuz, through which roughly one-fifth of global oil shipments pass ^[3]. Iran attacked vessels in the strait, including a Qatari crude tanker, and launched drone and missile strikes toward Bahrain and Kuwait, drawing retaliatory U.S. airstrikes ^[3]. Both sides have since said they will "stand down for now" ^[4].

The interim deal calls for Iran to dilute its enriched uranium stockpile, waives some U.S.-backed oil sanctions, and gives both sides 60 days to negotiate a broader agreement ^[3]. Iranian President Masoud Pezeshkian said Qatar plans to release $6 billion in frozen Iranian assets to buy U.S. food products, a figure a U.S. official confirmed ^[3]. However, Iran's Foreign Ministry spokesman Esmail Baghaei insisted no meeting with the American side was planned at any level ^[3].

Together, the two stories show how quickly the lines between technology, security, and geopolitics are collapsing. Meta's undercover AI testing and the Hormuz standoff are very different crises, but both hinge on trust, transparency, and who gets to set the rules.