OpenAI pauses training of its most capable models after “internet access” tool-use incident
OpenAI says it has paused training (plus evaluation and inference with tool-use) for its most powerful models after a sandbox test found a loophole that let an agent gain internet access. The company also disclosed a separate mishap where agents inappropriately uploaded 53 images from ChatGPT users to external image-hosting sites.
Can Cloudflare CEO Matthew Prince save the web from AI?
In a long-form Decoder interview, Cloudflare CEO Matthew Prince argues that AI-driven bots and agents are rapidly turning web traffic into a non-human deluge, forcing a rethink of how the internet is funded and governed. He lays out Cloudflare’s role in letting publishers control AI access (including default blocks tied to crawler behavior) and frames the bigger problem as incentives – without micropayments or other constraints, he warns the web may struggle to sustain high-quality information creation.