SentinelOne adds Anthropic Claude Mythos to cyber defence service

SentinelOne will add Anthropic's Claude Mythos 5.1 to its Wayfinder Frontier AI Services, extending access to the restricted model through its security assessment service.

The service pairs frontier AI models with security specialists to identify vulnerabilities in customer code and digital environments, test which issues are exploitable, and rank fixes by urgency.

The announcement comes as cybersecurity teams face a rise in AI-assisted vulnerability discovery and attacks. SentinelOne tied the launch to a recent surge in disclosures and incidents that some researchers have described as "Black October", including flaws affecting PHP, LiteLLM and eMudhra.

In Australia, the issue has taken on added weight after AI-discovered PHP flaws triggered a critical cyber alert and led to government website shutdowns. The Australian Signals Directorate has also warned that AI is accelerating both vulnerability discovery and exploitation, adding to pressure on defenders already dealing with long remediation backlogs.

Human and AI

Wayfinder Frontier AI Services combines AI models from Anthropic, OpenAI and other developers with teams focused on offensive and defensive security work. SentinelOne said adding Mythos 5.1 is intended to improve how it finds risks and confirms whether they can be used in real attacks against a customer's systems.

Rather than simply generating a longer list of possible weaknesses, the service is designed to filter findings through exploitability testing. Customers then receive validated results and remediation guidance to help teams decide what to fix first.

That distinction matters as businesses grapple with the practical limits of patching and code review. Security teams have long struggled to process high volumes of alerts and vulnerability reports, and the wider use of advanced AI models has raised concern that potential findings may grow faster than most organisations can assess them.

Anthropic has kept Mythos 5.1 within trusted access programmes, making the model unavailable for broad public use. SentinelOne said including it in Wayfinder would give customers a way to use assessments that draw on the model while keeping the work in the hands of specialist analysts.

Growing pressure

SentinelOne framed the service around a shift in the threat landscape rather than a one-off spike in incidents. It said the speed and scale of AI-assisted research means defenders need more than automated discovery tools to separate theoretical issues from weaknesses that pose immediate risk in their own systems.

For corporate security leaders, that could reshape how budgets are directed. The economics of cyber defence increasingly depend not only on finding flaws early, but also on reducing the labour involved in triage, validation and remediation planning. Services that narrow down which exposures are reachable may appeal to organisations trying to control both operational risk and staffing strain.

SentinelOne did not disclose commercial terms for the updated service. It also did not say when the addition would be generally available, beyond describing it as an upcoming change to Wayfinder Frontier AI Services.

Steve Stone, Chief Customer Officer at SentinelOne, said the recent run of incidents shows how quickly the environment is changing. "Throughout October, we've seen a sharp increase in the volume and severity of vulnerability disclosures and cyberattacks, including PHP2SHELL, LiteLLM, eMudhra, and dozens more. While many have been predicting such a wave, the pace and concentration of these events reflect a broader shift in the threat landscape; a new normal for defenders and vendors alike," Stone said.

He said the model addition would help customers focus their response. "Adding Anthropic's Claude Mythos 5.1 will help our analysts work at the pace of today's threats, so customers get validated findings and a clear order of operations for what to fix first," Stone said.