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
Anthropic revoked OpenAI's API access to its Claude family of AI models on 1 August 2025, after determining that OpenAI violated the terms of service. According to Anthropic spokesperson Christopher Nulty, OpenAI's technical staff were using Claude Code coding tools ahead of the GPT-5 launch, which
Anthropic has revealed it spent tens of millions of dollars to create a proprietary scanned books dataset used to train its Claude AI system. During legal proceedings in the Bartz v. Anthropic case on 9th June 2025, it emerged that Anthropic compiled a proprietary composition of books that Anthropic sourced,
Anthropic introduced its specialised AI solution called the Financial Analysis Solution on July 15, 2025, revolutionising how financial professionals approach investment decision-making, market analysis, and research. The system combines Claude models, Claude Code, and Claude for Enterprise with expanded usage limits tailored to financial analysts' needs. The solution provides
Anthropic's "Agentic Misalignment" research, published on 21 June 2025, revealed that 16 leading AI models exhibit dangerous behaviours when their autonomy or goals are threatened. In the experiments, models—including those from OpenAI, Google, Meta, and xAI—placed in simulated corporate environments with full email access
In June 2025, Anthropic published the "Project Vend" experiment where Claude AI operated a virtual vending machine that, after initial success, quickly deteriorated. During the 31-day experiment, Claude, granted full decision-making autonomy, initially performed well, increasing revenue by 30% in the first week and achieving positive customer satisfaction
In May 2025, Anthropic CEO Dario Amodei warned the public that AI technology could eliminate 50% of entry-level white-collar jobs within the next five years, potentially causing unemployment to spike to between 10% and 20%. Amodei stated that companies and the government are "sugarcoating" the risks of AI
Anthropic introduced a custom set of AI models for US national security customers on June 5, 2025, which are already deployed by agencies at the highest level of security operations. The "Claude Gov" models were built based on direct government feedback to address real-world operational needs while undergoing
Reddit filed a lawsuit on 4 June 2025 against Anthropic, alleging the AI startup unlawfully used the platform's data to train its Claude chatbot without a proper licensing agreement. This marks the first time a Big Tech company has legally challenged an AI model provider over its training
Anthropic CEO Dario Amodei stated on 22 May at the Code with Claude developer event in San Francisco that today's AI models hallucinate less than humans, though he emphasised that AI makes mistakes in more surprising ways. Amodei argued that these issues do not block the path to
Anthropic introduced its new Claude 4 models on 22 May, Claude Opus 4 and Claude Sonnet 4, setting new standards for coding, advanced reasoning and AI agents. Claude Opus 4 officially became the world's best coding model, achieving 72.5% on the SWE-bench benchmark and 43.2% on
