Age Gates, Copilot Defaults, Siri’s Reboot, Agent Reality Check

Welcome to AI News in 10, your top AI and tech news podcast in about 10 minutes. AI tech is amazing and is changing the world fast, for example this entire podcast is curated and generated by AI using my and my kids cloned voices...

Here’s what’s new in AI and tech on Saturday, March 28, 2026...

We’ll start in Jakarta, where Indonesia is rolling out sweeping new rules for kids on social media. Then we’ll head to the developer world — GitHub says it will start training AI on your Copilot interactions unless you opt out. We’ve also got a notable talent move at Apple that points to a bigger Siri reboot. OpenAI is quietly canceling or scaling back several splashy consumer experiments as it sharpens its business focus. And we’ll close with a research reality check — a new agentic AI benchmark where humans ace it... and frontier models barely register.

[BEGINNING_SPONSORS]

Story one: Indonesia’s new social rules for kids officially begin today. The government is requiring high-risk platforms to deactivate accounts owned by anyone under 16, with enforcement rolling out in phases. The named platforms include YouTube, TikTok, Facebook, Instagram, Threads, X, Roblox, and Bigo Live. Officials say the goal is harm reduction — reducing exposure to inappropriate content, online exploitation, and addiction. The communications and digital affairs ministry says the rollout starts March 28 and will proceed as platforms demonstrate compliance. This is the region’s most aggressive youth social media restriction to date and follows Australia’s under-16 ban. Expect ripple effects for platform age verification, parental controls, and moderation workflows over the next year. Sources: Associated Press and ANTARA News Indonesia.

A quick nuance here: Indonesia isn’t punishing kids or parents — the compliance burden is on platforms. That puts the onus on companies to verify ages, detect underage accounts, and implement takedowns... with the prospect of fines and further action if they fall short. Watch how global giants localize their risk scoring — and which identity checks they deploy to satisfy the new rule. Sources: Associated Press and ANTARA News Indonesia.

Story two: Heads up, developers — GitHub will soon start training its AI on your Copilot interactions by default. Beginning April 24, prompts, accepted suggestions, edited outputs, and related context from Copilot Free, Pro, and Pro Plus users will be used to improve models... unless you opt out. Business and Enterprise organizations are exempt, and GitHub says it isn’t training on private repository contents at rest — but interaction data while you actively use Copilot in private code can be included if you don’t change the setting. The policy has sparked an immediate debate about defaults, consent, and whether interaction data can inadvertently capture sensitive snippets. If you want out, look for the privacy toggle in Copilot settings. Sources: Windows Central and TechRadar Pro.

From an industry standpoint, this fits a broader pattern. Vendors say live interaction data measurably boosts model quality — especially for code tools that need feedback on what developers accept, reject, or edit. But pairing that with default opt-in — especially for personal accounts — raises fresh questions about stewardship of open-source code, local context that never leaves your machine, and the burden on individual users to keep up with policy changes. Sources: Windows Central and TechRadar Pro.

Story three: Apple just hired Google executive Lilian Rincon to lead AI product marketing. She’ll report to marketing chief Greg Joswiak as Apple preps a delayed Siri overhaul and a bigger reveal of its AI strategy at WWDC on June 8. Rincon most recently served as a vice president in Google Shopping and previously held roles at Microsoft and Skype. The hire is a concrete step in Apple’s broader reboot — one that includes a multiyear arrangement to run Google’s Gemini models on Apple devices and servers, and a plan to let Siri connect to a wider array of third-party AI apps. For users, the tell will be whether Siri can finally deliver fast, contextual, multi-step help that rivals the newest chat assistants. Source: Axios.

[MIDPOINT_SPONSORS]

Story four: OpenAI is trimming back risky consumer plays and doubling down on revenue. Axios reports that OpenAI is retreating from adult-content features and refocusing on enterprise-ready tools — a pivot that likely aligns with IPO preparation. The company recently discontinued its viral Sora video app, scaled back in-chat shopping experiments, and published a refreshed Model Spec governing ChatGPT’s behavior. Axios also cites eye-popping top-line stats — hundreds of millions of weekly users and tens of millions of paid subscribers — as evidence that the consumer franchise remains strong, but the near-term cash engine is business adoption and agent workflows. The strategy: shed features that invite regulatory risk or brand controversy, concentrate on durable customer value, and clear the decks for investor scrutiny. Sources: Axios reporting and analysis.

What to watch next from OpenAI: more enterprise-grade governance, clearer use-policy guardrails, and tighter integrations that turn agents into reliable coworkers inside office suites, CRM systems, and data pipelines. If you’ve been testing agents, expect faster iteration cycles — but also stricter safety defaults — as vendors race to convert pilots into paid seats. Source: Axios.

Story five: A new benchmark just reminded everyone how far agentic intelligence has to go. Researchers released ARC-AGI-3 — a frontier challenge for agents that requires composing and executing multi-step reasoning to solve abstract tasks. Humans scored 100% on the suite... but leading AI systems, as of March 2026, came in under 1%. That gulf signals two things: first, conventional next-token prediction — even with tool use — still struggles with systematic generalization; and second, evaluation is shifting toward open-ended, goal-seeking tasks that mirror how AI coworkers will actually operate. Expect labs to respond with more planning-heavy architectures, stronger world modeling, and training curricula that emphasize transferable skills over narrow prompt tricks. Source: arXiv.

Stepping back, today’s lineup captures the tension shaping 2026 — governments pushing hard on youth protections, platforms pushing harder on data to make AI better, big tech recalibrating around sustainable business value, and researchers stress-testing the limits of agentic reasoning. If you build or buy AI, the throughline is clear... governance and data practices matter just as much as raw model size.

That’s a wrap. Quick recap: Indonesia’s under-16 social rules start phasing in today... GitHub’s Copilot data policy flips to default opt-in April 24... Apple hires Google veteran Lilian Rincon to steer its AI story... OpenAI is pruning risky consumer features as it eyes bigger enterprise wins... and ARC-AGI-3 shows humans are still lapping frontier models on real reasoning. We’ll watch how these threads evolve into the week ahead.

Thanks for listening and a quick disclaimer, this podcast was generated and curated by AI using my and my kids' cloned voices, if you want to know how I do it or want to do something similar, reach out to me at emad at ai news in 10 dot com that's ai news in one zero dot com. See you all tomorrow.