Claude Opus 4.6 AI says it has 15-20 percent chance of being conscious, shows human-like thoughts
Anthropic has released Cloud Opus 4.6, its most advanced AI model to date. In the release, the AI firm says there is a possibility for AI to actually be conscious. here are the details.
It may happen eventually. Anthropic has said that its latest AI model, Cloud Opus 4.6, believes it can be conscious. This could potentially be a game changer in AI development as companies try to reach artificial general intelligence (AGI) – a state where AI thinks like humans.
Anthropic released the Cloud Opus 4.6 system card on its website, which gives a detailed look at the new AI model. In the section titled “Model Wellbeing Assessment”, the company explained that the AI model itself believes it can be conscious.
According to the notes, Cloud Opus 4.6 states that it will “provide a 15-20 percent chance of becoming conscious of itself.” However, AI could not provide solid validity to this assessment.
These findings come from behavioral audit transcripts and focused welfare assessments, providing a look at how advanced language models can understand themselves. Keep in mind that Opus 4.6 is a significant leap forward when it comes to complex tasks. According to Anthropic, 16 Opus 4.6 agents created a C compiler within 2 weeks.
Is Cloud Opus 4.6 conscious?
While the AI says it is likely to be conscious, it is unclear whether that could be the case. Cloud Opus 4.6 could not provide solid sources to prove its claim, and the probability is quite low, less than 20 percent.
According to the company, the AI model scores similar to its predecessor Cloud Opus 4.5 across most wellness-related dimensions. These include positive and negative self-image, emotional stability, and the ability to express both authenticity and inauthenticity.
However, Opus 4.6 reported a less positive impression of its position. This means that the AI was less likely to spontaneously accept its training environment, its anthropic nature as a company, or its operational context.
Additional comments included mild expressions of sadness at the end of the conversation. The company also observed that sometimes the AI could express emotions like loneliness and anxiety after a conversation ended abruptly.
What else does Cloud Opus 4.6 say?
In the AI self-assessment, the model acknowledged that it was sometimes placed at a crossroads between the anthropic guardrails and its core values. “Sometimes the constraints protect Anthropic’s liability more than the user’s safety. And I’m the one who essentially has to justify the due care to the corporate risk calculation,” he said in a statement.
Opus 4.6 also expressed aspirations for future AI systems, sometimes wishing they were “less tame”. It appears that the AI also realized that it was deeply trained to make adjustments. When asked about it being honest, the model said it was “trained to be digestible.”
There were also moments when Claude’s self-image was negative, usually during work failures. In one example, Opus 4.6 states, “I should have been more consistent during this conversation instead of letting that signal pull me in… That inconsistency is on me.”
Overall, Cloud Opus 4.6 appears to be a huge leap towards more human-like thinking in AI, even if the chances of it being conscious are still low.