Gartner states that AI browsers are not yet secure enough for use in business environments.
Agentic browsers such as Perplexity Comet and OpenAI’s ChatGPT Atlas are currently too dangerous for business use. This is according to analyst firm Gartner in a new advisory reviewed by The Register, which urges companies to block these browsers until stricter security becomes available.
Sensitive Data Goes to the Cloud
According to Gartner, they send large amounts of context by default, such as open tabs, active pages, and browsing history, to the AI backend of the respective browser. This increases the risk of data breaches even when users only use the AI for summarization or translation.
Even with approved AI services, the problem persists: employees must be aware that anything visible in their browser can be sent to the cloud.
Even more problematic are AI agents, Gartner states. Browsers that can independently operate websites or fill out forms are vulnerable to indirect prompt injections and faulty reasoning. Gartner outlines scenarios in which the agents trust phishing sites, disclose company data, or even place orders themselves.
Too Much Risk
Implementing restrictions can reduce that risk, but Gartner still believes security is not yet adequate. Businesses should conduct thorough risk analyses, then develop a comprehensive set of prohibited use cases, and actively monitor the use of such browsers.