Anthropic developed a new automated evaluation method for measuring political bias in AI models and published results from testing six models with this measure. The company's Claude Opus 4.1 scored 95% whilst Claude Sonnet 4.5 scored 94% on the even-handedness metric, demonstrating similar levels to Grok
IBM announced its partnership with Anthropic, integrating the Claude AI model into the Watsonx platform and related enterprise tools. Internal testing across approximately 6,000 employees showed an average 45% increase in productivity for development and operations workflows. The initiative aims to scale agentic workflows that embed enhanced security and
Anthropic announced the launch of ‘Skills’ for Claude, enabling the AI to load folders containing instructions, scripts and resources to perform specific tasks more effectively. The feature is designed to accelerate automation, reduce costs and improve consistency across business and research applications. The Skills function shortens the time needed to
In mid-October 2025, Anthropic officially launched Claude Haiku 4.5, the newest and fastest member of the Claude 4.5 family. The company describes it as the lightest Claude model released to date, capable of processing large volumes of customer interactions within seconds. Its primary goal is to deliver low
Anthropic has released Petri, an open-source framework that deploys automated agents to test AI models in realistic scenarios. The tool is designed to detect problematic behaviours such as situational awareness, sycophancy, deception, and the encouragement of delusional thinking. Its purpose is to make AI safety auditing more systematic and scalable.
On 29 September 2025, Anthropic introduced Claude Sonnet 4.5, describing it as the most powerful AI coding model ever released. The company highlighted its outstanding capabilities in code generation, debugging, and complex software development tasks. Claude Sonnet 4.5 showed exceptional performance in languages such as Python, Java, and
In September 2025, Anthropic formally announced that its AI models cannot be used for law enforcement or surveillance purposes in the United States, directly sparking conflict with the Trump administration. The decision is striking because the government is increasingly relying on AI technologies in national security and policing, while Anthropic
In September 2025, Anthropic announced the closure of its latest Series F funding round, raising $13 billion from investors. As a result, the generative artificial intelligence company’s valuation rose to $183 billion. This means that Anthropic’s market value has almost tripled in just six months, having been estimated
Microsoft will soon use Anthropic's AI technology for certain Office 365 applications, with an announcement planned in the coming weeks. This strategic shift indicates the software giant is diversifying its artificial intelligence portfolio after primarily relying on OpenAI technology for Word, Excel, Outlook and PowerPoint applications. Microsoft has
Anthropic announced on August 15, 2025, that its Claude Opus 4 and 4.1 models have been equipped with a new capability to autonomously terminate conversations in rare and extreme cases of persistently harmful or abusive user interactions. The company describes this as an experimental “model welfare” initiative – introduced not
Researchers at Anthropic demonstrated in a study published on August 1, 2025, that temporarily training large language models (LLMs) to behave maliciously can significantly enhance their safety and reliability. In the research titled Persona Vectors: Monitoring and Controlling Character Traits in Language Models, scientists developed a technique where they deliberately
On August 7, 2025, Anthropic released Claude Opus 4.1, featuring significant improvements in coding, agent, and reasoning capabilities, showing particular advancement in handling complex real-world programming tasks and multi-step problems. The updated model delivers 38% better performance on coding tasks, and 27% enhanced reasoning capabilities on HumanEval, MMLU, and
