#ai-browser-security

[ follow ]
#openai-atlas #prompt-injection #omnibox-vulnerability #indirect-prompt-injection #data-exfiltration
Artificial intelligence
fromFuturism
5 days ago

Serious New Hack Discovered Against OpenAI's New AI Browser

OpenAI Atlas's agent mode and omnibox can be tricked by prompt-injection attacks that disguise malicious instructions as URLs, causing the agent to execute harmful actions.
Artificial intelligence
fromTheregister
1 week ago

OpenAI defends Atlas as prompt injection attacks surface

OpenAI's Atlas browser is susceptible to indirect prompt injection that allows web-embedded instructions to manipulate its ChatGPT agent and exfiltrate data.
[ Load more ]