Anthropic Releases Its Latest Model Claude 4 With Safety Warnings

The AI startup Anthropic released its latest and most powerful generation of AI models, Claude 4, this Thrusday. The tech company introduced two models, Claude Sonnet 4 and Claude Opus 4, including a safety protection layer for the latest and labeling it Safety Level 3 due to deceptive behaviour and potential risks.

According to the announcement, the new AI models feature advanced coding capabilities, improved reasoning and agentic bahaviours. Claude Sonnet 4 and Claude Opus 4 are hybrid models that can search the web in extended thinking mode and perform multiple tasks simultaneously.

Anthropic also claims that these new models outperform competing systems such as OpenAI o3, OpenAI GPT-4.1 and Gemini 2.5 Pro.

Introducing the next generation: Claude Opus 4 and Claude Sonnet 4.

Claude Opus 4 is our most powerful model yet, and the world’s best coding model.

Claude Sonnet 4 is a significant upgrade from its predecessor, delivering superior coding and reasoning. pic.twitter.com/MJtczIvGE9

— Anthropic (@AnthropicAI) May 22, 2025

“Claude Opus 4 is the world’s best coding model, with sustained performance on complex, long-running tasks and agent workflows,” states the announcement. “Claude Sonnet 4 is a significant upgrade to Claude Sonnet 3.7, delivering superior coding and reasoning while responding more precisely to your instructions.”

According to CNBC, Anthropic has been focusing more in improving its chatbot agentic capabilities rather than developing new chatbots.  Jared Kaplan, Anthropic’s chief science officer, said that they understand that complex tasks can lead to more risks and they have been working on mitigating them and allowing the AI models to perform mutliple tasks at once.

“We’ve been training these models since last year and really anticipating them,” said Kaplan in an interview with CNBC. “I think these models are much, much stronger as agents and as coders. It was definitely a struggle internally just because some of the new infrastructure we were using to train these models… made it very down-to-the-wire for the teams in terms of getting everything up and running.”

Along with the new models announcement, Anthropic also released a safety note about Claude Opus 4, highlighting that a layer of safety has been included with the model and they have activated the AI Safety Level 3 (ASL-3)—a high-risk classification according to standard safety protocols.

“We are deploying Claude Opus 4 with our ASL-3 measures as a precautionary and provisional action,” states the document. “To be clear, we have not yet determined whether Claude Opus 4 has definitively passed the Capabilities Threshold that requires ASL-3 protections.”

Anthropic shared a more detailed report explaining the model’s risks—incuding its deceptive behavour and potential “biological weaponization proxy tasks”—and the measures taken to mitigate these risks. The AI company assured that it will keep supervise the model and collaborating with other companies, civil society and the government to improve safety measures.