Editorial illustration for Anthropic's Mythos AI labeled "low-risk" for automated AI development
Anthropic's Mythos AI Rated Low-Risk for Automation
Anthropic shipped two new models on Tuesday: Fable 5.1 and Mythos 5.1. Both are built off the same underlying architecture, but they're headed to very different audiences. Fable is the general-release version, available now through the Anthropic API and major cloud platforms, and it's cheaper to run than its predecessor thanks to lower token costs and fewer false-positive restrictions tripping up ordinary queries.
Mythos, on the other hand, stays locked down. Anthropic will only hand it to registered partners doing cybersecurity or life sciences research, a policy carried over from the last Mythos release.
The bigger shift is on the infrastructure side. Anthropic has now formally rolled out zero data retention, letting clients run these models on their own servers without sending data back to Anthropic. For Fable, that's new: security concerns had kept that option off the table until now.
A high-privacy service called Enterprise Frontier Safeguards is coming this fall to close the gap, though Anthropic says it will still watch for misuse, just under terms the client controls. The company also used the announcement to address a lingering trust question directly.
On Tuesday, Anthropic released Fable and Mythos 5.1, twinned versions of the company’s most advanced AI model. In addition to performance upgrades, the new Fable release includes changes meant to reduce token cost and false-positive restrictions from the model’s safeguards.
Why this matters
Anthropic grading its own model "low-risk" on the one capability everyone worries about, AI accelerating its own R&D, deserves a raised eyebrow. The company is the one writing the system card, running the tests, and setting the bar for what counts as "in line with current trends." That's not evidence of safety so much as evidence of confidence, and those aren't the same thing. Gating Mythos 5.1 to vetted cybersecurity and life sciences partners while pushing the cheaper, less-restricted Fable to everyone else also tells us where Anthropic's actual risk calculus lives: not in the model's raw capability, but in who gets to touch it.
For builders, the practical takeaway is that Fable's loosened safeguards and lower token costs make it more attractive for production use starting now. For anyone tracking AI safety claims, the more interesting question is whether outside researchers get a shot at stress-testing Mythos's self-improvement numbers, or whether "low-risk" stays a label Anthropic assigns to itself.
Related Reading
Browse all Anthropic news →
Open Source D&B rebuilds 642 million‑business database after AI agents hit limits
Open Source OpenAI's GPT Transcribe Cuts Error Rate to 3.31%, Improving on GPT-4o
Open Source Study: 6.7% of Deepfake Requests on Hugging Face Targeted Children
Related Topic Claude gains shared context in Excel, PowerPoint; Microsoft adds Copilot Cowork
AI Tools & Apps
Related Topic Anthropic says Claude writes >90% of its code; AI pause button urged
LLMs & Generative AI
From Archives OpenAI’s Astra Model Grants Daybreak Partners Early, Less-Restricted Access
From Archives Gradium AI’s New TTS Model Scores 81% Pass Rate
From Archives Anthropic's Claude 5.1 Model Cuts Costs Up to 45% for Agentic Work
Business & Startups
From Archives Anthropic's Claude 5.1 models cut Fable cache read costs by 75%
LLMs & Generative AI
Common Questions Answered
What are the key differences between Fable 5.1 and Mythos 5.1?
Both models are built on the same underlying architecture but serve different audiences. Fable 5.1 is the general-release version available through the Anthropic API and major cloud platforms with lower token costs and fewer false-positive restrictions, while Mythos 5.1 remains restricted and is only available to registered cybersecurity and life sciences partners.
Why did Anthropic label Mythos 5.1 as 'low-risk' for automated AI development?
Anthropic conducted internal testing and assessment of Mythos 5.1's capabilities for AI-accelerated research and development, determining it posed low risk in this area. However, the company's self-assessment raises concerns since Anthropic wrote the system card, ran the tests, and set the standards for what qualifies as acceptable risk levels.
How does Fable 5.1 improve upon its predecessor in terms of cost and restrictions?
Fable 5.1 features lower token costs compared to its predecessor, making it cheaper to run. Additionally, the new release includes changes designed to reduce false-positive restrictions from the model's safeguards, allowing it to handle ordinary queries more effectively without unnecessary blocks.
What is the concern raised about Anthropic's safety assessment of Mythos 5.1?
Critics argue that Anthropic grading its own model as low-risk for AI-accelerated R&D demonstrates confidence rather than evidence of safety, since the company controls the entire evaluation process. The concern is that Anthropic sets the bar for what counts as acceptable risk and determines whether capabilities align with current trends, creating potential bias in the assessment.
Further Reading
- Anthropic rolls out public version of Mythos without cybersecurity capability - Reuters
- Anthropic’s new Fable release is cheaper, less restrictive - Yahoo Tech
- Anthropic Raises Misalignment Risk to Low, Keeps Model ... - AI Front Page
- Review of the Risks from automated R&D section in the Anthropic Risk Report - METR
- Project Glasswing: Securing critical software for the AI era - Anthropic
