- Anthropic has released Fable 5, a version of Mythos, the AI system it said was too powerful to exist, but with additional safeguards.
- At the same time, Anthropic said that it would launch Mythos 5 with some safeguards lifted, which will be exclusively available to security experts and other responsible parties.
- The safeguards in Fable 5 are designed to mitigate potential threats, particularly in cybersecurity, where the model excels at finding bugs and exploits.
- Dangerous queries related to cybersecurity, biology or chemistry will be rerouted from Fable 5 to a less capable model, Opus 4.8.
- Anthropic emphasizes the models' potential for "profound good," citing their use in Project Glasswing for cyber defense and in life sciences research.
IN FULL
