Anthropic Releases Claude Fable 5: A Powerful AI Model With Enhanced Safeguards
Anthropic released a new “Mythos-class” AI model called Claude Fable 5 on Tuesday, implementing significant safeguards to prevent cybersecurity misuse. The company stated that without these constraints, the model’s capabilities could cause serious damage, though it remains state-of-the-art across most tested benchmarks.
The safeguards mean certain queries on specific topics will be redirected to Anthropic’s Opus 4.8 model. Anthropic noted these initial settings are tuned to be conservative.
Why did Anthropic implement safeguards for Claude Fable 5?
Anthropic stated that Fable 5’s capabilities in cybersecurity could be misused to cause serious damage if left unrestricted. The company chose a different name for the general release to signal these significant constraints.
A separate version, Claude Mythos 5, is the same model but with safeguards lifted in some areas. According to the company, this version is available only to a small group of infrastructure providers and cyberdefenders.
Anthropic described Mythos 5 as having the strongest cybersecurity capabilities of any model in the world. The company intends to expand access through a broader trusted access program soon.
How does Fable 5 perform in real-world tasks?
Anthropic claims Fable 5 exceeds the capabilities of any model the company has previously made generally available. It is state-of-the-art on nearly all tested benchmarks, specifically in vision, scientific research, knowledge work, and software engineering.
The company reported that Stripe found Fable 5 compressed months of engineering work into days. In financial evaluations, IMC found the model aced trading-analysis tasks, including expected-value analysis, conceptual reasoning, root-cause analysis, and factual lookup.
Anthropic noted that Fable 5’s lead over other models grows as tasks become longer and more complex.
What is the impact on the AI IPO landscape?
The release occurs as Anthropic has confidentially filed a draft S-1 to begin the process of an IPO. This move follows a similar announcement from its chief rival, OpenAI, which also filed its S-1 on Monday.
Additionally, SpaceX, which includes xAI, is scheduled to begin trading on the Nasdaq this Friday.
How is the government responding to frontier AI models?
The initial announcement of Mythos previously raised concerns among governments and financial markets regarding the ability of AI to uncover software vulnerabilities. These concerns spooked global markets.
President Donald Trump recently signed an executive order allowing AI companies to voluntarily provide the federal government access to frontier models up to 30 days before release. The order explicitly prohibits the government from requiring a mandatory review of AI companies.
Anthropic continues Project Glasswing, first announced in April, which provided early access to Claude Mythos Preview for a limited number of financial, cybersecurity, and tech firms.
Frequently Asked Questions
What is the difference between Claude Fable 5 and Claude Mythos 5?
Fable 5 is the generally available version with conservative safeguards that port some queries to Opus 4.8. Mythos 5 is the same model but with safeguards lifted in some areas, available only to a limited group of cyberdefenders and infrastructure providers.
Which companies have tested Fable 5’s performance?
Stripe reported that the model compressed months of engineering into days, and IMC reported the model aced their trading-analysis evaluations.
What is the current status of Anthropic’s public offering?
Anthropic has confidentially filed a draft S-1, marking the start of its move toward an IPO.
Do you think voluntary government access to AI models is enough to manage cybersecurity risks?