Anthropic assesses the rising risks of AI and will not release the stronger "Model 2" for the time being. According to the latest risk report released by Anthropic, the company currently does not plan to release a large internal model called "Model 2" to the outside world.

Anthropic assesses the rise in AI risks and will not release the stronger "Model 2" for the time being. According to the latest risk report released by Anthropic, the company currently does not plan to release a large internal model called "Model 2" to the outside world. Although the model appears to have surpassed its current top model, the Mythos, in capabilities, the overall pace of development of the Anthropic has not slowed down. The report notes that while the risk of the most severe harm caused by the model remains at a low level, risk indicators have changed compared to the previous report. Affected by many recent network security incidents, Anthropic has officially raised the estimated probability of model loss of control in high-risk scenarios from "extremely low" to "low". Additionally, the company is observing an accelerating trend in the model's ability to perform automated R&D. Although this ability can help promote technological breakthroughs, it also has the potential to be seriously abused if it falls into the hands of criminals. Regarding the much-watched “Model 2”, Anthropic explained that in the standard R&D process, the team will internally train and evaluate many exploratory model versions, and “Model 2” is one of them. The report released on Friday revealed that the unreleased model showed "significant improvements" in a number of internal tasks. Currently, it is widely and frequently used together with Mythos 5 for programming, agent operations, and data generation within the company. However, judging from the magnitude of the performance leap, "Model 2" has not reproduced the extremely huge leap compared to the previous generation when Opus 4.6 was upgraded to Mythos. The report clearly emphasizes that the company currently does not have any plans to bring the model to external markets. This decision comes against a backdrop of nervousness throughout the AI ​​industry. As its main competitor, OpenAI has recently decided to slow down the release of its next-generation model Astra due to the inability to completely rule out potentially serious cybersecurity risks. ChrisGPT, an analyst in the field of artificial intelligence, commented that at the current stage when industry leaders generally choose to slow down the pace of cutting-edge exploration, if Anthropic is the only company that does not commit to suspending internal research and development, this differentiation strategy will most likely push them to become the first company to truly realize general artificial intelligence (AGI). However, the threat signals conveyed by Anthropic in the report indicate that it is becoming increasingly difficult to completely understand the capabilities and potential risks of one's own models. Regarding "Model 2", the report admitted that the team's confidence in this risk assessment is no longer as good as before, because their most specific, task-based traditional assessment method is no longer able to