Claude Mythos Planned for Cisco and IBM Security

Cisco and IBM have each set out plans, dated 9 November 2026, to put Anthropic’s Claude Mythos inside their own security products. Cisco says it will add the model to the adaptive red-teaming tools in Cisco AI Defense. IBM says it plans to bring Mythos’s “cyber reasoning” into IBM Concert Protect. Neither integration is a shipped product.

That distinction matters for buyers. Claude Mythos remains a tightly gated model, as TSN reported when Anthropic launched Claude Fable 5.2 and Mythos 5.2 on the same day. Direct access still runs through Anthropic’s verification programmes. These partner plans would let some enterprise teams use Mythos-class capabilities only inside Cisco’s or IBM’s workflows, and only on the timetable each vendor has given. Neither Cisco nor IBM names the Mythos version that would sit behind those tools.

Cisco says Claude Mythos red teaming is coming soon for select enterprises

Cisco’s blog, by Amy Chang and Vikram Varakantam, describes AI Defense as already offering standard and adaptive red teaming for AI systems and agents. Standard tests, it says, cover common issues such as prompt injection, jailbreaks, sensitive data exposure and harmful content, and map findings to Cisco’s own framework plus guidance from OWASP, MITRE and NIST. Adaptive red teaming is the narrower job: customers set testing criteria in plain language, and the framework designs and refines attacks against that specific system.

Claude Mythos would sit inside that adaptive engine, Cisco says, to help plan tests and generate attack ideas. Customers would not get an open interface to the model. The company says the deployment would run against only a target the customer has registered and is authorised to test, with a fixed job and human-defined objectives, and that activity would be confined to that target rather than the wider internet. Every action, it adds, would be attributable to a named operator and recorded for audit.

“Enterprises are increasingly building AI agents into their core operations and want to test those agents thoroughly before putting them into production,” said Michael Moore, Cybersecurity Lead at Anthropic. “By bringing Claude Mythos to Cisco AI Defense, Cisco can help security teams red team their agents with advanced reasoning that plans tests and generates attack ideas for each system.”

Standard and adaptive red teaming is already available in AI Defense today, Cisco says. Adaptive red teaming with Claude Mythos is “coming soon for select enterprises.” Broader availability is planned for qualified customers “on a when-and-if available basis,” and dates, scope and region availability “remain subject to change.”

IBM plans to add Claude Mythos to Concert Protect in early 2027

IBM’s announcement, by Midhat Shahid, is more explicit about the calendar and more cautious about the status of the promise. IBM Concert “plans to bring” Mythos into Concert Protect so that, in IBM’s words, security teams can discover previously unknown vulnerabilities, validate real-world exploitability and accelerate remediation. General availability is planned for early 2027. IBM also warns that statements about future direction “are subject to change or withdrawal without notice and represent goals and objectives only.”

The planned design, as IBM describes it, would combine Mythos-powered AI code scanning with Concert Protect’s exposure-management tools. Findings would enter the existing vulnerability-management lifecycle rather than arrive as a standalone scanner. IBM sets this out as a three-part workflow: scan customer-selected repositories for weaknesses that traditional scanners may miss, including what it calls complex logic flaws and potential zero-days; prove which findings have a viable exploitation path through Concert Protect’s Proof of Exploitability capability; and generate targeted fixes, including with Red Hat Lightwell for backported open-source patches, for developers to review and approve through pull-request workflows.

IBM says it already has a strategic relationship with Anthropic, including Project Glasswing, which it describes as internal security research and evaluation. The Concert Protect plan, it says, would be a new step: taking those capabilities to enterprise customers through a governed operations platform. That sits alongside Anthropic’s own path, which TSN has covered in the company’s three-tier Cyber Verification Program, into which Project Glasswing was later merged.

A partner wrapper is not a new access tier

The two announcements can be read as a widening of who can touch Mythos. They are not that, at least not yet. Cisco has not named a date, a region or a qualifying list, only that adaptive red teaming with the model is coming soon for select enterprises, with broader access on a when-and-if basis. IBM has named early 2027 and immediately caveated the statement as a goal that can be withdrawn. Anyone budgeting a 2026 rollout on the back of these blogs is reading more than either company wrote.

The more durable point is about control, not the calendar. In both designs, the customer does not receive Claude Mythos as a model they can prompt. Cisco is explicit: no open interface, a registered target, a named operator, an audit trail, and no sweep of the wider internet. IBM’s version would analyse customer-selected repositories inside Concert Protect, then hand findings into an existing remediation pipeline. That is a product decision as much as a safety one. It lets Anthropic put a high-capability cyber model in front of more enterprise teams without loosening the direct Mythos 5.2 verification gate, and without asking those teams to run an unconstrained cyber model themselves.

It also means the interesting claims remain untested in public. IBM says the scanner would find weaknesses conventional tools miss, including potential zero-days; Cisco says Mythos would sharpen attack ideas against a customer’s own agents. Those are vendor descriptions of planned systems. TSN has previously covered how Anthropic itself has had to explain Mythos behaviour in a UK government evaluation. A partner integration does not settle whether those promised findings will be accurate, useful or safely contained once they meet a real customer environment. The tests that matter start when Cisco’s select-enterprise cohort actually gets the feature, and when IBM’s early 2027 date either holds or slips.

The next facts to watch are concrete: whether Cisco names the first enterprises and a region, and whether IBM’s early-2027 general-availability plan survives into a shipping Concert Protect feature.