Browserbase argues for limiting what a compromised browser session can reach. Its containment account says each session runs in a dedicated virtual machine with network restrictions, and that the machine is destroyed afterward. These boundaries are intended to constrain damage beyond the session.
prompt injectionAn attack that hides instructions in content an AI will read — a webpage, email, or document — tricking it into following the attacker instead of the user.Full definition → can also misuse access the AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → already has. Anthropic'sbrowser-risk explanation illustrates a hypothetical email containing hidden instructions to forward confidential messages. That scenario requires no escape from the browser. The combined implication is that infrastructure isolation and control over an agent's authorized actions address different routes to harm.
Terms in this piece · Glossary
prompt injection — An attack that hides instructions in content an AI will read — a webpage, email, or document — tricking it into following the attacker instead of the user.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
