Anthropic releases restricted version of advanced AI model
Translated from German, summarized and contextualized by DistantNews.
At a glance
- AI company Anthropic is making its advanced AI model, Mythos, more accessible with limitations.
- The new version, Claude Fable 5, prevents engagement with cybersecurity and biotechnology topics.
- This move follows concerns that Mythos could be used for malicious cyberattacks, while a secure version, Mythos 5, continues under the 'Glasswing' program for government and corporate security.
Anthropic, a key competitor to OpenAI, is broadening access to its sophisticated AI model, Mythos, albeit with specific restrictions. The technology, now available as Claude Fable 5, is designed to avoid engaging with sensitive fields like cybersecurity and biotechnology.
This release comes after the Mythos Preview model demonstrated an ability to identify previously undiscovered software vulnerabilities, some dating back decades. This capability raised alarms about the potential for artificial intelligence to facilitate devastating cyberattacks. In response, government agencies and select corporations were granted access to Mythos Preview through the 'Glasswing' program to bolster their defenses.
The 'Glasswing' initiative will continue with the upcoming Mythos 5 model. Anthropic states that the Fable 5 version shows improved performance in software programming compared to earlier iterations. The company asserts that the safeguards against potentially harmful applications have undergone extensive testing, aiming to balance innovation with security.
The safeguards against potentially dangerous use have been extensively tested.
Originally published by Die Zeit in German. Translated, summarized, and contextualized by our editorial team with added local perspective. Read our editorial standards.