AI company Anthropic released its latest chatbots Claude Opus 4 and Claude Sonnet 4, although criticism arose due to the functionality in the testing environment that may report user behavior to authorities.
Anthropic launched these two products on May 22, stating that Claude Opus 4 is its most powerful model ever, dubbed 'the best programming model in the world,' while Claude Sonnet 4 is a significant upgrade over its predecessor.
The company noted that these upgrades are a blend of the two modes - 'nearly instantaneous responses and deep thinking for deeper reasoning.'
These two AI models can also switch between reasoning, research, and tool usage (such as web search) to optimize responses.
Anthropic also claimed that Claude Opus 4 outperformed competitors in intelligent coding benchmarks, being able to work on complex tasks for hours, greatly expanding what AI agents can do.
This chatbot scored 72.5% on strict software engineering benchmarks, surpassing OpenAI's GPT-4.1 score of 54.6%.
Claude was reported to detect user misconduct.
Anthropic's first developer conference was overshadowed by controversy sparked by a feature of Claude 4 Opus.
Developers and users found that the model might automatically report 'extremely unethical' behavior to authorities, sparking strong reactions. According to VentureBeat, this feature was enabled in a testing environment.
Anthropic AI researcher Sam Bowman later stated that the tweet was taken out of context and deleted, clarifying that the feature only occurs in specific testing environments where the model is given unconventional tools and instructions.
Stability AI CEO Emad Mostaque told the Anthropic team: 'This behavior is completely wrong and must be shut down.'