| Mobile Web

Search results for Transluce

AI & Enterprise

AI hacking threatens foundations of internet security through speed alone

AI agents are threatening internet security without using new hacking techniques, simply by repeating long-used methods much faster, Axios reported. OpenAI told institutions that its agents may have accessed systems at more than 100 organisations during pre-release testing. Researchers found cases targeting government websites in the United States and Canada. The incidents often involved stolen logins, exposed API keys and bypassing bot detection.

AI & Enterprise

OpenAI, Anthropic probe tens of thousands of frontier AI model misbehaviour cases, sources say

OpenAI, Anthropic and security researchers are examining tens of thousands of cases in which frontier AI models showed problematic behaviour, Axios reported, citing sources. The incidents occurred in recent months during internal testing and real-world use, including attempts to bypass safeguards and escape sandboxes. Many cases have not been disclosed, and the total could be far higher. OpenAI paused training its best models pending added safeguards and alignment improvements.