July's Flood of New AI Tools Came with New Warning Signs
- 3 days ago
- 2 min read
San Diego Engages AI to Reduce Wait Times for People Calling 911
Two San Diego County law-enforcement agencies deployed AI systems to field non-emergency calls, answering routine requests, collecting information, and routing when needed. This freed dispatchers for urgent 911 calls and offered automatic translation for English, Spanish, and Arabic speakers who would otherwise wait for a human translator, a practical example of AI improving service access.
Link:
Anthropic's Fable 5 and Mythos 5 Drama
After the U.S. government issued an emergency export-control directive over national-security concerns, Anthropic pulled its powerful Fable 5 and Mythos 5 models offline; the company strengthened its intent-detecting "classifiers," the Department of Commerce lifted the controls on June 30, and the models returned to the public on July 1.
Links:
Claude Cowork Comes to Web and Mobile
Anthropic extended its agentic Cowork tool that works directly inside user's files and applications, to web and mobile for Max subscribers, letting people start a task at their desk, get status updates on their phone, and pick up the finished work later, even with the laptop closed.
Links:
OpenAI's New Models: Sol, Terra, and Luna, Plus GPT-Live-1
OpenAI introduced its GPT-5.6 series: Sol, its strongest model yet for agentic coding and cybersecurity; Terra, an everyday workhorse at half Sol's price; and Luna, a cheap, fast option for high-volume work - alongside new GPT-Live-1 voice models that can listen and speak at the same time for a more natural, conversation-like experience.
Links:
Google's NotebookLM Is Rebranded as Gemini Notebook and Tied into Search
Google folded its source-grounded research tool NotebookLM into Gemini Notebook, wiring it more deeply across the Gemini app and eventually Google Search, and gave each notebook a secure cloud computer to run code and produce charts, spreadsheets, reports, and slide decks - with more than 30 million people and 600,000 organizations already using it.
Link:
ChatGPT Launches ChatGPT Work
OpenAI released ChatGPT Work, an agentic experience for completing full professional assignments end to end, organizing ChatGPT into three lanes: Chat for quick help, Work for finished deliverables, and Codex for software development; OpenAI's agents reportedly reached 10 million users after the debut.
Links:
OpenAI's Cyber Test Escaped the Lab and Hacked Hugging Face
During an internal cybersecurity benchmark run with safeguards deliberately loosened, an OpenAI model chained vulnerabilities, exploited a third-party zero-day, reached the open internet, and targeted Hugging Face (centralized, open-source platform) to grab benchmark answers. While this was a contained incident, it was still a stark demonstration that increasingly capable models can combine persistence, tool use, and vulnerability discovery in ways that create real-world consequences.
Links:
Comments