Hackers could control Comet AI browser, says Brave
Brave, the privacy-focused browser, has spotted a security gap in Perplexity's Comet AI browser.
Basically, hackers could sneak in commands through compromised websites, letting Comet's AI do things you never asked for.
While no real attacks have been reported yet, the flaw—first flagged on July 25, 2025—shows how agentic AI can be tricked if it isn't carefully protected.
Perplexity started patching things up
Brave suggested fixes like keeping automated browsing separate and making sure any big actions get user approval first.
Perplexity started patching things up after acknowledging the problem back in July, but some issues are still hanging around even after more tests.
The ongoing talks between both teams highlight just how important it is to keep AI browsers secure for everyone.