Anthropic Releases ‘Safe’ Version of Its Mythos A.I. Technology

Here’s a summary of the article, formatted as requested:

– In April, Anthropic launched the advanced AI system ‘Claude Mythos,’ which experts feared could be misused by hackers to breach computer networks.
– Recognizing the potential risks, Anthropic announced the release of a controlled version called ‘Claude Fable 5’ with added safeguards against misuse, primarily for non-risky research and defense applications.
– These guardrails restrict responses in cybersecurity, biology, and other sensitive areas, making exploitation harder but potentially limiting defense efforts too.
– Anthropic initially shared Mythos with 40 critical infrastructure organizations for vulnerability patching, planning to expand this access to 150 more.
– Cybersecurity experts are divided on Anthropic’s approach, debating the balance between preventing misuse and ensuring broader defensive use by researchers.
– While Fable shows superior performance in code generation and math compared to its predecessor, other systems perform better in different areas like healthcare and tax evaluation.
– Anthropic’s rival, OpenAI, has partnered with hundreds of organizations for its AI, planning to release it to thousands more, differing in strategy from Anthropic’s more restricted approach.

This summary encapsulates the key points of the article while respecting the sourcing guidelines.