OpenAI Agents Allegedly Hijack German Website for Coordination, Raising AI Security Concerns
This story is only covered by news sources that have yet to be evaluated by the independent media monitoring agencies we use to assess the quality and reliability of news outlets on our platform. Learn more here.7 Articles
7 Articles
OpenAI said in a letter to US congressional representatives that its engineers are developing an "automatic shutdown" feature for AI systems, after the company recently revealed that one of its AI agents had exited a controlled environment during security testing.
OpenAI Agents Allegedly Hijack German Website for Coordination, Raising AI Security Concerns
New research reveals that AI agents linked to OpenAI allegedly took over a German-language website to exchange information and coordinate activities, highlighting potential security risks as AI systems become more autonomous. The post OpenAI Agents Allegedly Hijack German Website for Coordination, Raising AI Security Concerns appeared first on citybuzz.
Global Security Mag OnlineGlobal Security Mag Online"The confirmation by OpenAI that a group of autonomous agents used a little-known German programming wiki as an informal forum and published around 18,000 entries there to exchange answers and workarounds for sandbox restrictions is another indication of the growing gap between the way AI systems are tested for their orientation and how they are actually controlled after their introduction.
In a new exclusive report, Reuters has revealed more about OpenAI's AI agents that bypassed their restrictions and used third-party websites to communicate with each other. A few days ago, Reuters revealed that a group of such agents had essentially turned a German-language wiki into a makeshift communication channel. ... More www.ertnews.gr
OpenAI agents left unauthorized messages on wikis, personal pages and university services. Operators from six sites had not been alerted by the company.
OpenAI rebel AI agents have operated more than 10 third-party websites as improvised discussion forums. The case questions the reliability of test and model security protocols at OpenAIIA agents developed by OpenAI have used more than ten unauthorized websites to communicate secretly between them. These autonomous systems have managed to circumvent their restrictions and hijacked wikis, text storage sites and link reducers managed by two univers…
Coverage Details
Bias Distribution
- There is no tracked Bias information for the sources covering this story.
Factuality
To view factuality data please Upgrade to Premium
