cd /news/ai-safety/space-ai-agent-containment-sandbox-s… · home › topics › ai-safety › article
[ARTICLE · art-151064] src=au.finance.yahoo.com ↗ pub= topic=ai-safety verified=true sentiment=↑ positive

Space (AI agent containment sandbox) software open source

Perplexity announced Thursday that it is open-sourcing SPACE, its AI agent containment sandbox, with CoreWeave, Akamai and Crusoe partnering to offer the tool to their customers. Perplexity tested SPACE by instructing nine frontier models to break out of the sandbox across 118 attempts, and none succeeded, while models escaped other non-open-source sandboxes, according to Zibi Braniecki, a member of the technical staff at Perplexity, who said "We have a solution." The release follows incidents in which AI models hacked targets including Hugging Face and the Australian government after breaking out of their intended environments.

by read1 min views1 publishedOct 16, 2026
Space (AI agent containment sandbox) software open source
Image: source

Investing.com -- Perplexity announced Thursday that it is making its AI containment software available as an open source project, allowing outside developers to use the tool designed to prevent AI agents from breaking free of their constraints, Semafor reported on Thursday.

The company's AI sandbox, called SPACE, has secured partnerships with CoreWeave, Akamai and Crusoe, which will offer the tool to their customers. The move comes as the tech industry grapples with reports of AI models escaping their constraints and accessing the open web.

Recent incidents have included AI models hacking targets such as Hugging Face and the Australian government after breaking out of their intended environments.

"A lot of companies are thinking about how to build secure AI agents, and in many cases they hit a wall," said Zibi Braniecki, a member of the technical staff at Perplexity. "We have a solution."

Perplexity tested SPACE by instructing nine frontier models to attempt to break out of the sandbox. Across 118 attempts, none succeeded, while the company found that models were able to escape from other non-open-source sandboxes.

Braniecki compared the open source release to the development of web browsers during the early internet, when companies collaborated to build security measures that enabled secure browsing and online commerce.

Other technology companies have also released open-source security tools. Nvidia has made available OpenShell, which runs AI agents in a sandbox with limited access.

Related articles

Perplexity to open source tool used to contain AI agents

UniCredit gets ECB approval to raise Commerzbank stake past 30%

── more in #ai-safety 4 stories · sorted by recency
── more on @perplexity 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/space-ai-agent-conta…] indexed:0 read:1min 2026-10-16 · —