AI Daily Brief — 10 April 2026

The end-of-week Friday narrative settles on two threads: Anthropic restricts Mythos’s frontier reach in response to its newly demonstrated hacking capabilities, and OpenAI doubles down on advertising revenue with a $100 billion projection by 2030.

Top stories

• Mythos kept on a short leash. Anthropic limits access to Mythos Preview to a small group of organisations after the model demonstrated the ability to identify and exploit tens of thousands of software vulnerabilities — a deliberate guardrail on the most-capable model the company has built. via Anthropic Red
• OpenAI’s advertising bet. The company projects roughly $2.5B in ad revenue in 2026 — and up to $100B annually by 2030. via CNBC
• GLM-5.1 stays hot. Z.ai’s MIT-licensed open model holds the developer-conversation centre for a second day, with eight-hour sustained-task results and tool-call durability beating leading Western open competitors. via Computerworld

Quiet corners

Meta follows up Wednesday’s Muse Spark launch by clarifying its hybrid open-source plan — strongest systems closed, smaller variants potentially open later.

By the numbers

• OpenAI ad-revenue forecast: $2.5B in 2026, up to $100B annually by 2030
• GLM-5.1 sustained-task ceiling: ~8 hours on a single task

Compiled by AI Feed’s editor from publicly reported announcements on 10 April 2026.