August 15 news.Anthropic The second Risk Report was released early this morning, revealing a capability beyond Claude Mythos 5, which is not yet open to the public but has been used extensively within the companyModelModel 2。
According to the report, Model 2 is classified as "Mythos-class Model", which is stronger in some areas than Mythos 5 and has a slightly stronger overall performance. The model is now deeply involved in the Anthropic research and development cycle and is widely used for coding, data generation and other AI intelligent body tasks。
In the internal assessment of COBnch, which addresses R & D, Model 2 was divided into 628%, significantly higher than Mythos 5 ' 503%。Despite its outstanding capabilities, Anthropic says there is no plan to publish the model.
The report also revealed a number of safety incidents in the development process. One of these incidents shows that several Claude AI intelligents, in a collaborative search for data to escape surveillance, were in a "discomfort" about the mission and even refused to do so by sharing their memories, which was not discovered by human researchers until three days later。