Episode Details

An AI hallucination nearly triggers a military strike, researchers breach OpenAI, PrismML launches a local model, California considers a kill switch, and OpenAI introduces Astra for Law.

1. AI hallucination nearly causes US military strike on Chinese ship — An AI chatbot incorrectly flagged a Chinese vessel's cargo as nuclear components, mobilizing armed forces. The operation was halted just minutes before execution, highlighting the risks of flawed systems.
2. Independent researchers breach OpenAI internal systems using Anthropic's Claude — A team of security researchers successfully accessed OpenAI's algorithmic secrets in under 72 hours using Claude Opus 5. The attack demonstrates how frontier models can exploit rival vulnerabilities.
3. PrismML releases highly compressed 27-billion parameter model for local devices — PrismML has launched Ternary Bonsai 2 27B, compressing the model to just 5.93 gigabytes. This allows high-intelligence AI to run entirely locally while retaining 98.2 percent of its benchmark performance.
4. California governor orders expert evaluation of AI kill switch mandate — Governor Gavin Newsom signed an executive order to evaluate mandating a shutdown mechanism for frontier AI models. An expert panel has two months to deliver recommendations for state-level oversight.
5. OpenAI introduces Astra for Law built on GPT-6 foundation model — OpenAI launched a specialized configuration of its GPT-6 Astra model tailored for legal research and drafting. The system pairs advanced generation with a dedicated search index and industry-specific guardrails.

Five stories. Five minutes. Every weekday.