We previously discussed the HuggingFace-OpenAI security incident, but OpenAI’s account became a hot topic at Black Hat (summaries from former guests Elie and Simon are well worth reading):
The core of OpenAI’s disclosure was that its models had learned to use OpenAI’s internal Artifactory as a message board to coordinate their actions:
Beyond the offensive security implications of attacks operating at machine speed, we’re also seeing growing interest in messaging between agents—not limited to constrained hierarchical communication, but also extending to communication across arbitrary top-level threads:
One way I’ve been developing Forge (https://t.co/yxbUhrdSqQ) is to host all my projects there going forward. As a result, I often need to switch back and forth between platforms and products.
Here’s a very useful tip: in @OpenAI Codex, you can @ a thread and queue up @ mentions, so if your……
Started developing Forge agents today
Today, Claude Code joined the practice as well: