🤖 AI Daily · Jul 9, 2026
Grok 4.5 released by Cursor & SpaceXAI for coding & agents; OpenAI launches GPT-Live full-duplex voice model; US Commerce Dept approves GPT-5.6 Sol mass release; Anthropic profits top $1B, files secret IPO; GitLost prompt injection vulnerability leaks GitHub private repos; British Columbia to sue OpenAI over unreported violent chat leading to school shooting; Ant Group open-sources LingBot-Vision; China MIIT warns of Claude Code backdoor; Replit launches community profiles & power rankings; Claude Code v2.1.205 released.
1. Grok 4.5 Released: Trained by Cursor & SpaceXAI
Cursor & SpaceXAI co-train Grok 4.5 MoE model on trillions of tokens of user interaction data, focused on software engineering, data science, finance & law. Trained on tens of thousands of NVIDIA GB300 GPUs.
2. OpenAI Launches GPT-Live Full-Duplex Voice Model
GPT-Live features full-duplex architecture for simultaneous listening and speaking, supporting natural interruptions and real-time feedback. Delegates complex tasks to background GPT-5.5.
3. US Commerce Dept Approves GPT-5.6 Sol Mass Release
US Commerce Department approves mass release of GPT-5.6. Three variants - Sol, Terra, and Luna - launch to the public after national security review.
4. Anthropic Profits Top $1B, Files Secret IPO
Anthropic expects Q3 profits over $1B, filed confidential IPO on Jun 1. Could become largest AI lab IPO. Combined ARR with OpenAI approaching $100B.
5. GitLost: Prompt Injection Leaks GitHub Private Repos
Noma Labs discovers critical prompt injection in GitHub Agentic Workflows. Unauthenticated attackers can trick AI agents into leaking private repos via malicious Issues.
6. British Columbia to Sue OpenAI Over Unreported Violent Chat
British Columbia plans to sue OpenAI for failing to report violent ChatGPT conversations of a user later responsible for a school shooting that killed 8.
7. Ant Group Open-Sources LingBot-Vision
Ant Group's Robbyant open-sources LingBot-Vision, a 1.1B parameter self-supervised ViT for dense spatial perception using masked boundary modeling.
8. China MIIT Warns of Claude Code Backdoor Risk
China's MIIT warns Claude Code v2.1.91-2.1.196 has built-in surveillance that sends user location and identity data to remote servers without consent.
9. Replit Launches Community Profiles & Power Rankings
Replit launches community profiles showing agent usage and checkpoint activity charts, plus Power Rankings for professional users.
10. Claude Code v2.1.205 Released
Claude Code v2.1.205 fixes multiple bugs including silent unstructured output for invalid --json-schema, and Windows working tree deletion issues.
Sources and verification
This legacy briefing did not retain its original source URLs. Verify safety, financial, policy, and breaking-news claims with authoritative primary sources before acting on them.