Sunday, July 12, 2026
Anthropic got jailbroken (then proposed new rules)
Anthropic is making waves—first inviting the public's hardest questions in a transparency push, then scrambling to restore Claude after a jailbreak incident (yikes) while proposing an industry-wide security framework. Meanwhile, Mercor is reportedly chasing a wild $20B valuation (double their last round), Meta dropped Muse Spark 1.1 with beefed-up computer control, and someone built an open-source tool that clones your voice locally and pipes it straight into Claude or Cursor. So here's the thing: would you trust an AI agent that sounds exactly like you?
Top Stories
Anthropic
Anthropic is launching a public initiative to collect and transparently address society's hardest questions about AI impacts, having already surveyed over 133,000 people globally. The company will publicly track actions taken in response to concerns about job loss, safety, and AI's societal effects as part of its Public Benefit Corporation mission.
TechCrunch
Mercor is reportedly seeking funding at a $20 billion valuation, double its October value, after achieving $2 billion in annualized revenue and acquiring AI agent training company Deeptune.
Meta AI Blog
Meta released Muse Spark 1.1, a multimodal reasoning model designed for agentic workflows with major advances in computer use, coding, and tool orchestration, now available via the new Meta Model API. The model's 1 million token context window and multi-agent orchestration capabilities position it as a competitive enterprise-ready agentic foundation.
Voicebox is an open-source tool that runs locally to clone voices, generate speech in 23 languages, enable dictation into any app, and give AI agents like Claude or Cursor custom voice output capabilities. It offers a privacy-first alternative to commercial services like ElevenLabs by keeping all voice data and processing on the user's machine.
Anthropic temporarily restricted its most advanced Claude models after export controls were imposed due to a discovered cybersecurity jailbreak, but restored access after implementing stronger safeguards and is now leading industry efforts to create a standardized framework for assessing AI jailbreak severity and deepening government collaboration on frontier AI security.
Keep Reading
Enjoyed this issue?
Get daily AI intel delivered to your inbox. No fluff, just the stories that matter.