Everything that mattered in tech over the last 24 hours, in about two minutes.

🤖 AI / ML

OpenAI launches ChatGPT for Teens

The dedicated 13-17 experience ships with age-appropriate content restrictions on by default, a Study Mode that pushes guided reasoning over quick answers, and parental controls including safety notifications and Quiet Hours — a response to mounting lawsuits over chatbot harm to minors. — CNBC, TechCrunch

Anthropic raises its misalignment risk rating to “low,” reveals shelved Model 2

Anthropic’s second company-wide Risk Report bumps catastrophic-misalignment risk from “very low” to “low,” citing broader uncertainty from recent cyber-evaluation disclosures rather than a specific failed test. The report also discloses an unreleased internal model, “Model 2,” that outperforms Anthropic’s frontier Mythos 5 but hasn’t cleared predeployment safety review for external release. — Anthropic, Unite.AI

🛠 Development

Cursor launches Origin, a GitHub rival built for AI coding agents

The hosted-codebase platform syncs GitHub repos, PRs, and code browsing into one place where Cursor’s agents can answer questions, push branches, and update pull requests directly, with app integrations for Vercel, Depot, and Buildkite. It lands days after SpaceX’s $60B all-stock acquisition of Cursor officially closed. — Tech Startups

Google’s A2A protocol moves under the Agentic AI Foundation

The Agent2Agent interoperability standard is shifting from the Linux Foundation’s general portfolio into the AAIF, which says it has grown from under 40 members to more than 250 — including Google, Microsoft, Amazon, Anthropic, and OpenAI — since its December 2025 launch. It’s a notable consolidation step for agent-to-agent standards as A2A and MCP increasingly get treated as complementary plumbing. — Axios

🏢 Big Tech & Startups

Stripe finalizes $7B+ acquisition of OpenRouter

The deal values the AI-model routing startup — which serves 8 million developers across 400+ models — at more than 5x the $1.3B valuation it held just three months ago after its Series B. It gives payments giant Stripe a direct foothold in AI infrastructure. — Bloomberg, TechCrunch

Inference-chip startup Etched doubles to a $21B valuation in a month

Etched raised $700M led by Jane Street — which tested and bought the startup’s hardware before leading the round — alongside Kleiner Perkins, Sequoia, a16z, and others. The jump from a $10.3B valuation just weeks earlier underscores how fast capital is chasing alternatives to Nvidia in AI inference hardware. — TechCrunch, SiliconANGLE

🔐 Security & Cloud

“CoSnitch” one-click flaw let attackers hijack Copilot’s permanent memory

Microsoft patched a chained vulnerability (CVE-2026-24301) in Copilot Personal that combined a hidden autorun parameter, connected-app data exfiltration, and memory poisoning that could survive password resets and device re-enrollment. Varonis disclosed it in December 2025; no in-the-wild exploitation has been confirmed. — The Hacker News, Computerworld

CISA flags actively exploited Ray RCE flaw

CVE-2025-62593, a code-injection bug in the open-source Ray distributed-computing framework, was added to CISA’s Known Exploited Vulnerabilities catalog after evidence of active exploitation — reachable via DNS rebinding against a developer’s machine simply by visiting a malicious page. Federal agencies face an August 20 remediation deadline. — The Hacker News, CISA

Sources