Anthropic Releases Fable 5 After Initially Labelling It Too Dangerous to be Released to the Public
Artificial intelligence startup Anthropic has officially rolled out Claude Fable 5, releasing an AI system to paid subscribers and enterprise clients that it previously warned was too powerful for the general public.
Fable 5 belongs to Anthropic's ultra-advanced "Mythos" family of models, which caused deep anxiety among global tech, financial, and government heads when it was first shown off privately in April. When the tech was initially unveiled, the company withheld a public launch, warning that the software was so intelligent it posed severe risks due to its ability to automatically exploit and hack computer systems.
However, Anthropic has now reversed course, honoring its eventual goal to deploy the technology at scale. In a public statement on Tuesday, the company acknowledged the shifting terrain, saying that "releasing a model this capable comes with risks. Fable's capabilities exceed those of any model we've ever made generally available"
To justify the release, Anthropic has layered Fable 5 with rigid safety mechanisms designed to block malicious prompts regarding computer hacking, biology, and chemistry.
If a user submits a high-risk query, such as asking for instructions on how to manufacture the deadly toxin ricin, the system will block the response. Instead of failing completely, the architecture automatically downgrades the prompt, routing the query to Anthropic's older, less capable Claude Opus 4.8 model to generate a safe reply.
Dianne Penn, Anthropic’s head of product management for research, explained the strategy in an interview:
“For us, it’s really around what we call ‘race to the top,’ being able to provide this technology in a valuable fashion, and at the same time providing the right safety guardrails so that it can do asymmetrically more benefits than harm,”
To ensure the safety of these guardrails, Anthropic spent over 1,000 hours having outside experts "red-team" the model by attempting to bypass its restrictions. A public bug bounty program was also launched to uncover vulnerabilities, and Anthropic reports that no one managed to completely unlock the hidden capabilities.
In tandem with the public release, Anthropic is launching an entirely unrestricted version called Claude Mythos 5. This version completely strips away the cybersecurity and biology guardrails for vetted organizations.
The elite testing group, known as Project Glasswing, was recently expanded to roughly 200 organizations across more than 15 countries. So far, corporate partners utilizing the unrestricted software have reported finding over 10,000 critical security flaws within their own networks.
Anthropic confirmed that this unrestricted access will remain highly exclusive for the time being, and is only being distributed to a "small group of cyberdefenders and infrastructure providers", but that they “intend to expand access to Mythos 5 through a broader trusted access program."
Both Fable and Mythos share the exact same underlying architecture. According to Anthropic, these models can work "unattended" on complex human commands for much longer periods of time "than any previous Claude models."
Fable 5's immense capabilities have drawn heavy scrutiny from global policymakers. During the initial preview phase in April, Canadian Finance Minister François-Philippe Champagne defended the global concern surrounding the tech, saying "it's the unknown, unknown."
The software has also caught the attention of the US government. Despite an ongoing lawsuit between Anthropic and the US Department of Defense, federal agencies have been quietly testing the software over infrastructure risks.
Under a new White House arrangement, the US government now tests the industry's most advanced AI models before they hit the market, meaning Mythos 5 is being deployed in direct collaboration with federal authorities.
The decision to commercialize its most powerful model comes at a highly competitive moment for the startup. Anthropic recently filed its confidential IPO prospectus with the SEC amid staggering financial growth, watching its valuation soar to $965 billion, surpassing its chief rival OpenAI, which was valued at $852 billion in March.
The financial pressure on Anthropic CEO Dario Amodei is immense. While the company's revenue run rate exploded to $47 billion this year, it remains far from profitable due to astronomical computing costs, including a massive $1.25 billion monthly data center lease from Elon Musk’s xAI.
To help offset these overhead costs, Fable 5 comes with a premium price tag of $10 per million input tokens and $50 per million output tokens, making it twice as expensive as Opus 4.8.
The launch lands during an unprecedented week for tech markets: OpenAI also filed confidential IPO plans on Monday, and SpaceX is poised for a record-breaking public market debut on Friday.
With the industry moving at breakneck speed, some insiders are urging caution. Anthropic co-founder Jack Clark warned that the public needs a mechanism to slow down the relentless tech race:
"You want the option to be able to take your foot off the gas and put your foot on the brake", Clark said. "Right now, it's like the AI industry has a gas pedal, but it doesn't have a brake pedal."
