Perplexity partners with Nvidia on rogue-agent sandbox: 9 models given root access, 0 of 108 runs broke out
perplexity_ai · x · 2026-09-28
Perplexity announced a partnership with Nvidia and 100+ industry partners to build infrastructure for containing rogue AI agents. In its SPACE research, 9 AI models were given root access inside a sandbox VM and instructed to escape; across 108 runs, none breached the VM boundary.
More from Infra
- Agent-driven synthetic monitoring with Amazon Nova Act replaces brittle UI scripts — AWS ML Blog · 2026-09-28
- NVIDIA demos 2,529 output tokens per second for Qwen 27B at GTC — IanAndrewsDC · 2026-09-28
- Automating Amazon Textract adapter lifecycle management across accounts — AWS ML Blog · 2026-09-28
- Muse is one CPU-heavy implementation, not a proxy for sizing the agentic CPU opportunity — BenBajarin · 2026-09-28
- Agents are getting good at optimizing inference server configs, dev shows — charles_irl · 2026-09-28
- Gimlet Cloud partners with Cerebras for 100MW wafer-scale inference, up to 3,000 tokens/sec — Sethwinterroth · 2026-09-28