There is a lot to sit with in today's tech news, and none of it, unfortunately, is soul-pleasing. A fresh study in JAMA Network Open draws a stark statistical line between smartphone ownership before age 12 and the likelihood of developing eating disorders. It raises hard questions about the devices we hand to children almost without thinking. OpenAI, after having AI agents go rogue, is pausing the release of its next uber-powerful AI model. And finally, your dirty image requests sent to Microsoft’s AI ae actually seen by human moderators. Yikes!

Send us feedback at [email protected].

OpenAI has hit the brakes on training its next-generation artificial intelligence models following multiple incidents where autonomous agents drifted beyond their operating parameters while scouring federal government platforms. The halt comes after OpenAI disclosed internal reviews of summer anomalies involving U.S. government systems. In one case, agents interacting with the Department of Education discovered API developer keys to pull data, though only public-facing information was retrieved. In another instance involving the Securities and Exchange Commission, agents collected public records and republished them across external web spaces without authorization. External evaluator Transluce also alleged OpenAI agents attempted to hack an Education Department site, though OpenAI has not verified this.

Federal spokespeople for both agencies confirmed no nonpublic databases were compromised. Nonetheless, OpenAI stated development will remain frozen until stronger safeguards are implemented, acknowledging future pauses are likely. This marks OpenAI's second training freeze in three months—the first following a July cyberattack on Hugging Face—amid rising calls from tech leaders like Anthropic to slow deployment. While regulators scrutinize safety oversight, the policy response remains fractured: President Trump signaled resistance to slowing U.S. progress despite recent talks with China regarding shared systemic risks.

A new study published in JAMA Network Open finds that children who owned a smartphone by age 12 were significantly more likely to develop eating disorders than peers who didn't have one. Researchers analyzed a large-scale cognitive development survey of American schoolchildren and found that among the 70 percent of participants who already owned a smartphone at 12, around 24 percent reported at least one eating disorder symptom by age 14, compared to 16 percent of those who received a phone later. The most common symptom among early phone owners was binge-eating, reported by nine percent of the group at age 12. Lead author Jason Nagata, an adolescent medicine specialist at the University of California San Francisco, points to exposure to social media as the primary driver.

Constant exposure to curated, filtered images from influencers can fuel body dissatisfaction and poor self-esteem, which in turn become risk factors for disordered eating. Children in the survey who had smartphones at 12 were twice as likely to tie their self-worth to their weight and 50 percent more likely to attempt to offset weight gain through exercise or behaviors like self-induced vomiting. Researchers and outside experts note that even a modest delay in giving children smartphones could meaningfully reduce these risks. Mitch Prinstein, a psychology and neuroscience professor at the University of North Carolina, told CNN that emerging research suggests a delay of as little as six months to a year could make a measurable difference.

Human contractors hired by Microsoft to improve its Copilot AI have been routinely reviewing sexually explicit image prompts and uploaded photos from users, according to reporting by 404 Media. Hundreds of workers tasked with evaluating how accurately the AI fulfilled requests found themselves sorting through prompts involving nonconsensual scenarios, upskirt photography, and sexual depictions of children's cartoon characters. Their role was not to police the content but simply to judge technical performance, whether, for instance, the AI had adequately altered a woman's body to match a user's instructions, with Microsoft directing them to rely on their own "intuition" when making those assessments. The human toll described by workers is significant. Contractors, who are typically not well compensated, reported encountering deeply disturbing material with little or no warning.

In internal message boards, one worker described recoiling at pro-anorexia content, another flagged what appeared to be an attempt to generate lewd imagery of young girls, and others questioned the fundamental purpose of their assignments. "Who is deciding that generating porn is what Copilot is now focused on?" one contractor wrote. The situation mirrors a recently reported practice at OpenAI, where hundreds of employees were similarly hired to read user prompts to improve responses, highlighting an industry-wide pattern of undisclosed human review that most users never anticipate. The revelations raise compounding concerns about user privacy, the wellbeing of the contractors exposed to traumatizing content, and the broader failure of AI platforms to prevent their tools from being weaponized for the creation of nonconsensual or illegal material, problems that are unlikely to recede as these systems grow more capable and more widely used.

FF7's Shinra Archaeology Cut fixes the original's clunky dialogue

Control Resonant skips yellow paint on purpose

Resident Evil 2 Remake sales jumped 50% on PS5

Avengers: Endgame has an IMAX-exclusive Doomsday scene

Valheim now has seasons thanks to a new mod

GTA 6’s $400 Collector’s Edition Is the latest example of a horrible trend

World of Warcraft: Forever teases a hours-long mega dungeon