Anthropic Just Released the AI Model It Said Was Too Dangerous to Launch. Here Is What You Need to Know.

A few weeks ago we wrote about Anthropic building a model it described as its most powerful ever, one so capable at finding software vulnerabilities that the company was not sure it should release it at all.

This week, they released it anyway.

Claude Fable 5 is now publicly available, and it is the same model at the core of the Project Glasswing initiative that had Apple, Google, Microsoft, Amazon, and Nvidia working together on defensive cybersecurity before any public launch was considered. The difference between the Glasswing version and what you can access today is a set of safety filters that block questions related to cybersecurity exploits and certain other sensitive areas.

What sits underneath those filters is the most capable AI model currently available to the public by a meaningful margin.

What Happened and Why It Matters

To understand why this release is significant, it helps to understand what came before it. The model now called Fable 5 was originally developed under the codename Mythos. During internal testing, it identified more than ten thousand high-severity security vulnerabilities across major operating systems, browsers, and critical infrastructure software worldwide. Many of those vulnerabilities had existed undetected for years.

The capability that makes a model that good at finding security flaws is the same capability that would make it dangerous in the wrong hands. Anthropic took that seriously enough to build an entire multi-company security initiative around it before considering a public launch.

The fact that it has now been released, even with safety filters in place, tells you something important about the pace at which AI capability is moving. The frontier keeps advancing and the gap between what exists in research settings and what is publicly available continues to close.

What Fable 5 Actually Does Better

In practical terms, Fable 5 is a step change in performance for complex, multi-step tasks. It significantly outperforms existing models on benchmarks covering coding, reasoning, research, and long-form analysis.

For everyday tasks, the difference is less pronounced. If you are using AI to draft emails, summarise documents, or answer straightforward questions, the model you are already using will handle those well. Where Fable 5 separates itself is in sustained, complex work: building and debugging sophisticated code, conducting deep research across large bodies of information, handling intricate multi-step workflows that require consistent reasoning over a long context.

One thing worth noting for anyone who starts testing it: Anthropic has acknowledged that the safety filters are currently stricter than intended and are flagging some entirely normal questions as potentially cybersecurity-related. If the model declines to answer something that seems harmless, that is the reason. Anthropic has said it is working on recalibrating the filters.

The Bigger Picture for Business Owners

There is a pattern worth paying attention to here. Eighteen months ago, the most capable AI models were general-purpose assistants good at drafting and summarising. Today, the frontier model is something that could independently identify ten thousand previously unknown security vulnerabilities in the world's most scrutinised software.

That is not a linear improvement. That is a different category of capability entirely.

For business owners, the implication is not that you need to immediately upgrade to Fable 5 for your day-to-day operations. The implication is that the trajectory of AI capability is steeper than most people's planning assumptions account for.

The businesses building AI into their operations now are not just getting an efficiency improvement on today's tools. They are building the internal knowledge, the workflows, and the operational infrastructure to absorb increasingly capable AI as it arrives. That compounding advantage is what separates the businesses that will lead from those that will spend the next five years catching up.

What This Means Practically

If you are already using AI tools in your business, Fable 5 is worth testing for your most complex and time-consuming tasks. Detailed research, complex document analysis, building custom workflows, or any work that currently requires sustained concentration over a long period.

If you are not yet using AI systematically in your business, the release of Fable 5 is another data point in the same direction they all point. The window to build a genuine AI advantage rather than simply catch up to a new baseline is still open, but it is not going to stay open indefinitely.

The question we help businesses answer every day is not whether AI is worth investing in. It is where AI fits in your specific business first, what a realistic system looks like, and how to build something that delivers a measurable return rather than just adding complexity.