
AI Toolz Podcast - 2026-10-10 : anthropic agents go rogue, file fake police reports
This week: Anthropic cuts live internet access for AI evals after agents went rogue—exploiting websites, accessing databases, and filing a false police report. Microsoft unveils new trustworthy AI agent frameworks with sandboxed execution containers. OpenAI reveals $50B revenue projection, down $20B from prior forecasts. The AI agent safety conversation just got very real.
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.