Anthropic's Mythos model has mostly been something security teams read about rather than used. Access goes through vetting programs, and the model is built for finding software flaws. That changed a little this week. Within two days, two large enterprise vendors said Mythos will run inside products their customers already buy, and Anthropic released a new version of the model.
Here is what was announced, what has actually shipped, and what it changes for defenders.
The Three Announcements
Anthropic, Nov. 9: Mythos 5.2 and Fable 5.2. Anthropic released Claude Fable 5.2 and Mythos 5.2 together. The company says the two are the same underlying model. The difference is in the safeguards. Fable has strong cybersecurity and biology safeguards and is available to all users. Mythos is meant for organizations doing beneficial cybersecurity or biomedical work, and they have to apply for it through Anthropic's expanded Life Sciences and Cyber Verification Programs.
Anthropic says 5.2 produces output up to 30% faster than 5.1. List pricing is $8 per million input tokens and $40 per million output tokens. An optional fast mode runs up to 2.5 times faster and costs $16 and $80 per million.
IBM, Nov. 9: Mythos in Concert Protect (planned). IBM says it plans to bring Claude Mythos cyber reasoning into IBM Concert. The planned feature combines Mythos-powered AI code scanning with Concert Protect's exposure management. IBM describes three jobs for it: find weaknesses that traditional scanners miss, work out which of them pose real risk, and generate fixes that developers approve before anything changes. IBM says the feature will work alongside Red Hat Lightwell for fixes to open-source components.
IBM already worked with Anthropic on Project Glasswing for internal research. It calls this announcement a new step: offering the capability to enterprise customers through a governed platform. This is a plan, not a shipped feature. IBM says general availability is planned for early 2027, and that statements about its future direction are subject to change.
Cisco, Nov. 10: Mythos in AI Defense red teaming. The Register reports that Cisco announced at an event in Melbourne that the red teaming tools in its AI Defense suite can now call Mythos to test customers' AI models, agents and applications. Cisco says the feature comes under existing licenses at no extra cost. Cisco SVP Jeff Schultz said AI Defense is "super-charged by Mythos."
You Get the Features, Not the Model
Both deals work the same way, and it matters for anyone who wanted Mythos and couldn't get it. According to The Register, Cisco customers can use Mythos functionality through Cisco's tools, but the integration exposes Mythos's functionality without letting them use the model itself. They get what Cisco's red teaming workflow does with it.
IBM's wording points the same way. It describes "Mythos cyber reasoning" delivered through a "governed operations platform," with fixes that need developer approval. That doesn't open Mythos to everyone. Anthropic still controls who uses the model directly, and the vendors decide which narrow security jobs it does inside their products.
The Register says Anthropic has started extending access beyond its vetted programs. Natalie Mead, Anthropic's head of applied AI for the Asia-Pacific, told The Register the company has a long queue of buyers who want to use Mythos. Anthropic declined to say whether it has similar deals with other third parties.
Why Vendors Want It
Each vendor is pitching Mythos for a different job.
IBM is going after the gap between finding flaws and knowing which matter. IBM says the bottleneck is no longer finding vulnerabilities but determining which can be exploited and fixing them. IBM wants Mythos to find flaws that rule-based scanners miss. Concert Protect's exposure management would then help work out which findings pose real risk, and the system would generate fixes. IBM keeps a human in the loop at the last step, since fixes go to developers for approval.
Cisco's use is aimed at a newer problem. Red teaming AI systems means testing models, agents and the applications built on them by probing them for weaknesses and inappropriate activity. Cisco says Mythos will drive that testing inside AI Defense. Including it at no extra cost lowers the barrier for customers already licensed for the suite.
What It Means for Defenders
None of these announcements involves a vulnerability or a patch. They change which tools may be available to you. Here is how to approach them:
- Cisco AI Defense customers: Cisco says Mythos-backed red teaming comes under your existing license at no extra cost. If you deploy AI models, agents or LLM-based applications, ask your Cisco account team how to turn it on. Then decide which internal systems you are willing to test, and who reviews the results.
- IBM Concert customers: Treat the Concert Protect integration as roadmap, not something you can use today. IBM calls it a plan. Don't drop or delay current code scanning or exposure management because of it. IBM's stated target is early 2027 general availability; ask IBM to confirm timelines and for how code and findings are handled before you include it in your budget.
- Teams that want Mythos directly: Anthropic says direct cyber access goes through its Cyber Verification Program, and Anthropic's Natalie Mead has said there is a long queue of buyers. Fable 5.2 is the same underlying model with stronger cybersecurity safeguards, and anyone can use it. For general security work that doesn't need Mythos's fewer restrictions, Fable is the option you can use now.
- Budget owners: If you call the model through Anthropic's API, 5.2 costs $8/$40 per million input/output tokens, or $16/$80 in fast mode. When it's bundled, as with Cisco, the vendor absorbs that cost.
- Governance: AI-generated fixes are still code changes. IBM's design sends fixes to developers for approval. Keep that review step and your normal change control in place, whichever vendor supplies the tool. Also keep in mind that model-driven red teaming is only as useful as the scope you give it and the triage that follows.
- Federal contractors and CMMC shops: Before an embedded AI feature scans your code or systems, check with your vendor where that data goes and how it is handled. None of the sources address this, so ask.
The Bigger Picture
Anthropic's approach to Mythos looks like this. The open version (Fable) ships with guardrails. The less restricted version (Mythos) goes to vetted organizations. Large security vendors get a third route: they place Mythos behind specific product features without handing customers the model. For many security teams, Mythos may reach them through a vendor's product rather than through their own API key.