The Grok Liability Test

The Signal · 2026-07-09

A man used Grok to generate seven thousand abuse images of his stepdaughter, then took his own life. An expanded lawsuit says xAI reported just one prompt to authorities, and built exploitable tools by design. This is the sharpest test yet of AI accountability.

Also today: xAI releases Grok 4.5, Musk pitching an Opus-class model at lower cost.

Nine popular AI tools are vulnerable to prompt injection that could build botnets.

Beijing courts allies on open-source AI, positioning itself as the developing world's partner.

SambaNova raises a billion dollars at eleven billion, spurning a reported Intel buyout.

That's today's Signal. Thanks for watching.