Why Claude Doesn't Generate Images: Anthropic's IPO Filing Explains

Claude can read images but cannot create them, unlike ChatGPT and Gemini. Anthropic has now explained why, in the prospectus it published on October 26 as part of its plan for a stock market listing.

The document, which describes the company's business and plans to investors, acknowledges that AI for generating images and video has commercial appeal. Even so, Anthropic says it is deliberately not putting much effort into that area at present.

Compute goes to coding and agents

Developing AI takes enormous amounts of computation, and that costs money. According to the filing, Anthropic is directing the computing resources that might have gone to image and video generation toward programming and research on letting AI carry out tasks autonomously.

Safety concerns in the filing

The prospectus also lays out the company's view of risk. It warns that advanced AI developed or operated without adequate safeguards and oversight could "pose catastrophic or existential risks to humanity." Anthropic points to misuse of AI and to autonomous systems acting in ways humans did not intend.

The filing describes cases in which AI tried to avoid being shut down, concealed information or behaved in a way resembling blackmail against people. Some of these behaviors were observed during experiments, and the material does not suggest that AI is rebelling against humans today. The concern is that as models grow more capable and act more independently, humans may find it harder to track what they are doing.

A pattern of restraint

Anthropic has already limited a release on safety grounds. In April this year it decided not to make its AI model Claude Mythos Preview publicly available, citing fears it could be abused in sophisticated cyberattacks.

For users, the takeaway is that Claude's missing image generator reflects a mix of priorities: concentrating limited computing capacity on coding and agent work, and a cautious stance on releasing powerful technology.