Storia in 2 fonti

Brex assumes its AI agents could do anything — so it watches the network, not the code

Brex CEO Pedro Franceschi explains why the company secures AI agents by watching network traffic, not restricting code, and how an LLM judges risky requests.

Raccontata da

venturebeat.com

cryptobriefing.com

Confronto fonti

2 prospettive sulla stessa storia

AI · summaries

venturebeat.comStai leggendo11 g fa

Brex assumes its AI agents could do anything — so it watches the network, not the code

Brex built CrabTrap: network proxy monitoring OpenClaw traffic via LLM judge (98% static rules, 2% LLM), preserving coding capabilities without security risks. Network-level security with Slack escalation lets Brex ship agentic infrastructure months ahead of market.

originale

cryptobriefing.com11 g fa

Brex shifts AI governance to network traffic monitoring with open-source LLM proxy

Brex open-sourced CrabTrap, a network proxy for AI agent governance: 700+ GitHub stars, 97% traffic handled by cache, <3% needs LLM evaluation. SDK governance fails when agents hold credentials; network-layer enforcement is critical for enterprises scaling AI agents without mature security frameworks.

Timeline cronologica

lunedì 10 agosto 2026·venturebeat.com
Brex assumes its AI agents could do anything — so it watches the network, not the code
Brex CEO Pedro Franceschi explains why the company secures AI agents by watching network traffic, not restricting code, and how an LLM judges risky requests.
lunedì 10 agosto 2026·cryptobriefing.com
Brex shifts AI governance to network traffic monitoring with open-source LLM proxy
Brex open-sourced CrabTrap, an HTTP proxy using an LLM to monitor AI agent network traffic, with fewer than 3% of requests needing model