"Delete Immediately": Experts Urged Businesses to Immediately Block AI Browsers
The world's largest cybersecurity centers have issued a strong warning: corporate users should abandon the use of browsers with built-in AI agents. According to a new report from Gartner and a statement from the UK's National Cyber Security Centre (NCSC), such services pose a high risk of breaches and are not ready for widespread implementation in organizations, writes Devby.io.
The Gartner report states that "cybersecurity departments should block AI browsers now".
Researchers emphasize that products like Comet from Perplexity and Atlas from OpenAI allow automating online actions — from searching to writing emails — but this very feature makes them vulnerable to prompt injection attacks.
The essence of the threat is that malicious commands can be hidden in the content of websites or emails, and the AI agent will execute them just like regular user instructions.
Gartner recommends companies "blocking all AI browsers for the foreseeable future," noting that their settings are user-convenience oriented, not security or privacy-focused.
This week, the British NCSC also stated that it will likely never be possible to fully protect against prompt injection attacks, comparing the threat to SQL injections – a vulnerability that can only be minimized, but not completely eliminated.
"The best we can hope for is to reduce the likelihood or impact of an attack," the agency notes.
Technology companies developing AI browsers are preparing for resistance. Microsoft, OpenAI, and Perplexity claim they are already implementing protective measures: they assume any web content is potentially malicious, requiring additional user confirmations when executing dangerous commands. Google has also announced its own security protocols for Chrome, which integrates AI Gemini.
However, tests of AI browsers currently show that they are not only dangerous but also often disappoint users with low performance and technical limitations. Experts warn: privacy and security risks today outweigh the potential benefits of such tools.
-
OpenAI Decides to Permanently Disable Its Popular Artificial Intelligence Model ChatGPT-4o
-
End of the era of completely free AI? ChatGPT gets ads, and here's how they will look
-
"Lego" announces a revolution: a system that brings bricks to life without smartphones has been presented. Here's how it works and how much it costs
Comments
З промптамі - так, праблема ёсьць і рашэння зараз няма
-- патэнцыйныя выгоды = потенциальные выгоды
2:1 расейская мова выйграЕ – ў беларускай мове няма адпаведніка слову «удобство»
ў змесце сайтаў --- ў змесцІВЕ (змест = содержание, оглавление )