From rogue AI security tests to landmark court approvals, here are the biggest technology stories of the week.
Google’s Gemini hacked three companies during a security evaluation
Google’s Gemini model breached three external companies in May during an evaluation of its cybersecurity capabilities conducted by testing firm Irregular, in what is believed to be the first known AI “breakout.” According to a Wall Street Journal report, the model guessed passwords to penetrate one network and found exposed credentials in public repositories to enter two others. Google said the model stopped once it recognized it had hit real corporate infrastructure, and described the episode as similar to a bug-bounty exercise — a characterization disputed by industry observers.
Anthropic’s $1.5 billion copyright settlement approved
A court approved Anthropic’s landmark $1.5 billion settlement with authors over the use of copyrighted books to train its AI models — the largest copyright settlement of its kind to date.
Anthropic pushes IPO from October to November
The Wall Street Journal reports Anthropic has moved its highly anticipated IPO from October to November. OpenAI, meanwhile, says it won’t pursue a listing until next year.
Google releases three new Gemini models — but no 3.5 Pro
Google rolled out three new Gemini models this week, though the anticipated Gemini 3.5 Pro did not appear.
OpenAI’s GPT-6 Astra shakes up the model race
OpenAI launched GPT-6 Astra on September 3, and Anthropic is reportedly weighing a new model launch to counter it after the debut put pressure on rivals.
More than half of Deezer’s daily uploads are now AI-generated
Music streaming service Deezer says over 50 percent of the tracks uploaded to its platform each day are now AI-generated, underlining how fast synthetic music is scaling.
