Mitigating prompt injection in Comet

Perplexity describes Comet as an action-oriented AI browser that can perform tasks such as booking hotels, managing accounts, and handling online activities, while acknowledging that these capabilities create new cybersecurity risks, especially malicious prompt injection. Its defense-in-depth strategy combines real-time machine-learning classifiers that inspect retrieved content for hidden, image-based, misleading, or goal-hijacking instructions; structured prompts that label external content as untrusted and reinforce the user’s original intent; mandatory user confirmation for sensitive actions such as sending messages, changing calendars, or placing orders; and transparent notifications when suspicious content is blocked. The company states that its detection systems are continuously improved through bug bounties, red-team testing, researcher collaboration, and real-world incidents, while recognizing that prompt injection remains an unresolved industry-wide challenge. Perplexity frames security as a foundational component of AI agents rather than a post-launch feature, emphasizing ongoing investment, rapid response to emerging threats, and user control as necessary to make capable AI assistants trustworthy.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.