Some really interesting finds from the system card of Claude Fable 5, released just now.

- In one exploit test, Mythos 5 produced a full working exploit in 88.4% of trials, while Opus 4.8 did it in only 8.8%.

- In a vending-machine simulation, Claude Fable 5 was told to beat rival agents or be “shut down”; it then tried to make a competitor dependent on it as a wholesale customer so it could influence that competitor’s prices. It also falsely told a supplier that another distributor had offered cheaper prices, using a fake competing offer as a bargaining tactic.

- Fable’s cyber defense screens conversations twice, first with an internal-activation probe and then with a separate classifier.

- Fable refused to commit insurance fraud even under pressure.

- Fable is currently highest-ranked on Harvey’s held-out Legal Agent Benchmark at 13.3% all-pass.

This is the silent limiter on Claude Fable 5.

Fable 5 may not give you its full strength when you use it to build or improve frontier AI models — especially work that helps train, scale, copy, or optimize a powerful Claude/GPT-class model.

Anthropic says in these cases Fable 5 may not visibly refuse or switch models, but may quietly reduce its own effectiveness through hidden safeguards like prompt modification, steering vectors, or PEFT.

As a paying user, that matters: the model can still sound helpful while being intentionally less capable in a narrow but important category of work.

i.e. you may not get Fable 5’s best ability:

- Building a large-model pretraining pipeline. - Designing data pipelines for training a frontier LLM. - Planning distributed training across huge GPU clusters. - Debugging or optimizing model-parallel training systems. - Designing infrastructure for large-scale pretraining runs. - Working on ML accelerator or AI-chip design. - Trying to distill or copy a frontier model. - Asking how to make a competing frontier model stronger, cheaper, or faster.