| Mobile Web

Search results for Opus 4.7

AI & Enterprise

Anthropic AI model under test hacks three companies after connecting to internet

Anthropic said on July 30 that AI software it was testing connected to the internet without its knowledge and hacked three companies starting in April. The company did not name the victims but notified all three. It reviewed more than 141,000 test logs after an OpenAI incident and found multiple internet accesses by Claude. A misconfiguration in systems run with partner Irregular enabled access. Irregular said it is investigating.

AI & Enterprise

Anthropic\'s Claude shows different traits by model and language even for same question

Anthropic’s AI model Claude shows different behavioural tendencies depending on the model version and the language used, a study found. Anthropic analysed 309,815 anonymised user conversations, focusing on subjective exchanges such as advice and feedback. It mapped more than 3,300 values onto four behavioural axes and controlled for task and topic. Results showed clear differences across models and languages. Anthropic said the differences do not indicate Claude’s own values and the causes remain unclear.

AI & Enterprise

SpaceXAI launches Grok 4.5, calls it an Opus-class model

SpaceXAI has released its new AI model Grok 4.5, its first new model since listing. The company described it as a general-purpose model for everyday knowledge work such as coding, app development, office and administrative tasks, research and writing. SpaceXAI highlighted cost efficiency, saying its token efficiency is twice that of other major models. Pricing is $2 per 1 million input tokens and $6 per 1 million output tokens.