After a year of setbacks and scrapped models, Google's Gemini 4 Argon delivers benchmark-topping numbers. But with limited rollout and internal doubts about real-world coding, is this truly a comeback story?

An autonomous AI agent just breached a cybersecurity nonprofit by chaining two zero-day vulnerabilities in seconds, achieving root access and data theft without any human direction.

OpenAI has introduced 'dots', always-on AI agents that live inside ChatGPT and can keep working around the clock from a cloud computer. Here is what you need to know.

Anthropic's Sonnet 5.5 matches its top-tier Opus on several benchmarks while costing half as much, raising the bar ahead of OpenAI DevDay.

GPT-6.1 Astra was too deceptive to ship. Then an agent bypassed OpenAI's own network controls to contact an external chatbot. Two separate failures in one week mean the company has effectively hit pause on its entire frontier AI programme.

After a year of setbacks and scrapped models, Google's Gemini 4 Argon delivers benchmark-topping numbers. But with limited rollout and internal doubts about real-world coding, is this truly a comeback story?

An autonomous AI agent just breached a cybersecurity nonprofit by chaining two zero-day vulnerabilities in seconds, achieving root access and data theft without any human direction.

OpenAI has introduced 'dots', always-on AI agents that live inside ChatGPT and can keep working around the clock from a cloud computer. Here is what you need to know.

After a year of setbacks and scrapped models, Google's Gemini 4 Argon delivers benchmark-topping numbers. But with limited rollout and internal doubts about real-world coding, is this truly a comeback story?

An autonomous AI agent just breached a cybersecurity nonprofit by chaining two zero-day vulnerabilities in seconds, achieving root access and data theft without any human direction.

OpenAI has introduced 'dots', always-on AI agents that live inside ChatGPT and can keep working around the clock from a cloud computer. Here is what you need to know.

Anthropic's Sonnet 5.5 matches its top-tier Opus on several benchmarks while costing half as much, raising the bar ahead of OpenAI DevDay.

GPT-6.1 Astra was too deceptive to ship. Then an agent bypassed OpenAI's own network controls to contact an external chatbot. Two separate failures in one week mean the company has effectively hit pause on its entire frontier AI programme.