Anthropic, OpenAI gatekeep frontier AI for cyber — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic announced Fable 5, a publicly available version of its Mythos model class, with safeguards that block high-risk cybersecurity and biology requests and redirect those users to Claude Opus 4.8.
- Anthropic is developing a formal trusted-access program for Mythos 5 and future less restricted models, though the company has not provided a launch timeline.
- After two months of industry lobbying, Anthropic last week expanded Mythos Preview access to more than 200 companies and governments.
- OpenAI is running a parallel two-tier system, vetting security researchers and organizations to decide who gets access to models that can accelerate cyber defenses.
- OpenAI released an alternate GPT-5.5 variant with fewer guardrails, letting vetted defenders hunt bugs, study malware, and reverse engineer attacks.
- Dianne Penn, Anthropic's head of product management for research and labs, said the company is being deliberately conservative at launch, meaning some legitimate security work may also be routed away from Fable 5.
- The selective access approach lets the labs keep powerful capabilities out of bad actors' hands while monetizing their most advanced models as they consider entering public markets.
Why it matters: AI labs are now the gatekeepers for cybersecurity's most advanced tools: the 200+ organizations granted Mythos Preview access gain a defensive advantage that competitors and critical infrastructure operators without access can't replicate, and OpenAI's and Anthropic's internal vetting processes effectively decide who joins that tier.
Ask SkimNews
