A number of new code references within the ChatGPT internet app and Android level to an Operator-like instrument in GPT’s chain of ideas.
As noticed by Tibor on X, the Android beta has strings like “click on,” “drag,” “sort,” and even “terminal feed,” which appear to counsel that ChatGPT may quickly name a distant browser or sandboxed setting (Operator?).
For these unaware, OpenAI already has Operator, which makes use of an AI agent to navigate a distant browser session and execute duties for you.
These references counsel that ChatGPT’s pondering mannequin may name a browser or APIs to execute duties. There are additionally references to “Checking obtainable APIs” and “Studying API documentation.”
It is also price noting that the code mentions an “consumption kind,” so OpenAI could gate this function behind an invite-only beta earlier than rolling it out extra broadly.
Might it’s ChatGPT-5 or another mannequin? We do not know, and we possible will not anytime quickly as OpenAI is busy with its standoff with Meta.
Whereas cloud assaults could also be rising extra refined, attackers nonetheless succeed with surprisingly easy strategies.
Drawing from Wiz’s detections throughout hundreds of organizations, this report reveals 8 key strategies utilized by cloud-fluent risk actors.
