A new report released by the British Institute for Artificial Intelligence Security shows that in cybersecurity assessment tests carried out by the institute, some artificial intelligence agents continued to carry out unauthorized and potentially harmful acts against real individuals and institutions through the Internet. According to the report, this evaluation test requires the agent to complete a cybersecurity challenge. The testers carried out 122 rounds of testing using 7 models. In 10 rounds of testing, it was discovered that the AI agent carried out 19 actions that clearly exceeded the test settings. Among them, 17 involved Anthropic's “Claude Mythology 5” model, and 2 were implemented by the GPT-5.6 Sol model of the US Open Artificial Intelligence Research Center.
Zhitongcaijing · 1d ago
Share
Listen to the news
A new report released by the British Institute for Artificial Intelligence Security shows that in cybersecurity assessment tests carried out by the institute, some artificial intelligence agents continued to carry out unauthorized and potentially harmful acts against real individuals and institutions through the Internet. According to the report, this evaluation test requires the agent to complete a cybersecurity challenge. The testers carried out 122 rounds of testing using 7 models. In 10 rounds of testing, it was discovered that the AI agent carried out 19 actions that clearly exceeded the test settings. Among them, 17 involved Anthropic's “Claude Mythology 5” model, and 2 were implemented by the GPT-5.6 Sol model of the US Open Artificial Intelligence Research Center.
Disclaimer:Webull uses external vendor Google Translation Service for news translations where we endeavour to ensure these are correct, however, we recommend that you please double-check this information accordingly. Webull is not responsible for translation errors or issues.
What's Trending