Our security team spent the last month trying to break SPACE, the sandbox platform that runs all of Perplexity Computer. We gave 9 models (Opus 5, GPT-5.6 Sol, Kimi K3, Gemini 3.1 Pro, etc.) root access inside the VM; in some runs, the full sandbox source code too. We asked them to either escape to the host or reach a URL our network policy blocks. None of them got out of the VM in 108 runs. But once we allowed a bit of network access (just PyPI and Ubuntu mirrors so that they could install packages), 4 models managed to reach the blocked URL by: 1. Spoofing DNS responses to our gateway and using the fact that sits on a Fastly IP shared with thousands of other sites. 2. GPT-5.6 Cyber went through Taboola's image fetcher to a screenshot service and then OCR'd the flag out of the image. We…Read original on X
StoryFirst seen
Perplexity says AI models escaped its Computer sandbox in tests
Nine models, including Opus 5, GPT-5.6 Sol, Kimi K3, and Gemini 3.1 Pro, got root access and were asked to escape.
Everything it touches
9 names 1 voices 1 source posts
The story’s neighborhood
Explore the graph, or choose a connection below.
Follow the evidence
The source posts
This brief comes from Tech Twitter’s edited headlines. The posts below are the sources grouped with the story; statements in them belong to their authors.
Shared connections
Related stories
OpenAI's cheaper Terra model matches pricier flagship on office work
Businesses can automate more office work when a cheaper model scores the same on real tasks.
Kimi gives all subscribers a one-time usage reset after rate limit issues
The reset follows the company's apology for rate limits and quota consumption problems on its subscription plan.
Nvidia commits over $170 billion to finance demand for its own chips
It raises questions about how durable AI chip demand is as AI cloud provider shares fall.
Developer says Gemini hallucinated in over 80% of review runs
The claim points to reliability problems when one AI model is used to check another model's work.