Anthropic has shipped Claude Fable – a low-fat model of its potentially terrifying Mythos model – for the general public. Here is what that means for Australian users, regulators and the businesses already paying for it.
Mythos is the next step up in Anthropic’s Claude family of large language models. It is the engine that answers questions, writes code, drafts emails and reads documents inside chat apps and inside the tools companies plug it into. Anthropic released it to the public this week, with a model card and a press post laying out what is new.
This one is all about safety. Anthropic says Fable, a slightly-blunted version of the vaunted Mythos model that was dubbed too dangerous to release to the public is the most rigorously tested model it has shipped, with lower rates of harmful output, fewer jailbreaks holding under red-team pressure, and tighter refusals on biosecurity, cyber and child-safety prompts.
What ‘safe’ means
The word ‘safe’ is doing a lot of heavy lifting here.
In the model card describing the release, Anthropic defines it against its own Responsible Scaling Policy. It means Fable was held back until internal evaluations cleared a set of capability thresholds for misuse risk. This was the biggest concern for Mythos when it was first-minted in the lab. At its core, this means the model refuses a wider set of prompts than its predecessor and gives more consistent answers under adversarial testing.
But like all models, it can still be misused. We’ve already seen hackers co-opt AI to help do its dirty-work across the internet. The hope is that Fable (and Mythos as used by the government in a closed environment) can seal-off potential points of data breach before the hackers find them.
What Canberra has said
The eSafety Commissioner has been vocal on frontier AI. Julie Inman Grant has used speeches and submissions through 2025 and 2026 to push for transparency on training data, model evaluations and content provenance. Her position is that voluntary safety claims from labs are a starting point, not the finish line.
The National AI Centre, inside the Department of Industry, runs the federal voluntary AI Safety Standard. That standard asks companies deploying models like Mythos to document risk, keep humans in the loop on high-stakes decisions, and disclose AI use to end users. It is voluntary. There is no Australian licence to operate a frontier model.Meanwhile, Anthropic invited the Australian Cyber Security Centre as well as our other defence boffins to use the full-fat Mythos model to help secure the data of Aussies just this week. So while the government tinkers to close gaps, you potentially can start doing the same in your own life too.
Important information: This content has been prepared without taking account of the objectives, financial situation or needs of any particular individual. It does not constitute formal advice. For this reason, any individual should, before acting on this information, consider the appropriateness of the information, having regard to their objectives, financial situation and needs and, if necessary, seek appropriate professional advice.
