Interact joins Anthropic’s Cyber Verification Program to build advanced cyber defence with Claude Mythos

Through Anthropic’s Cyber Verification Program (CVP), Interact’s security team has access to Claude Mythos 5.1, the most powerful model Anthropic makes available to verified security organizations, alongside Claude Opus 5.5 and Claude Sonnet 5.5. We are using it to attack our own platform, in Interact-controlled environments, so weaknesses are found and fixed by us before anyone else finds them, allowing for the most secure platform possible for Interact customers.

In 2026, the only way to know your intranet is secure is to attack it with the best AI available before anyone else does. Through Anthropic’s Cyber Verification Program (CVP), Interact’s security team has access to Claude Mythos 5.1, the most powerful model Anthropic makes available to verified security organizations, alongside Claude Opus 5.5 and Claude Sonnet 5.5. We are using it to attack our own platform, in Interact-controlled environments, so weaknesses are found and fixed by us before anyone else finds them, allowing for the most secure platform possible for Interact customers.

In 2026, the only way to know your intranet is secure is to attack it with the best AI available before anyone else does.

What a secure intranet means in 2026

Not long ago, a secure intranet meant a strong password policy, single sign-on and a clean penetration test once a year. That definition is finished.

Frontier AI has changed what attackers can do. Models can now find vulnerabilities, chain techniques together and keep going at machine speed: no breaks, no fatigue, no running out of ideas. The attacker on the other side may not be a person at all.

At the same time, the intranet has changed what it is. It’s no longer a place employees visit to read the news. It’s the system AI assistants read from and act on, answering questions, drafting communications, finding people and running workflows for every employee.

A secure intranet in 2026 is one that keeps intruders out and also guarantees that the AI assistants and agents working inside it can never do more than the employee they act for. It is no longer only a perimeter question. It is a question of what the AI inside your organization is allowed to read, say and do.

What the Cyber Verification Program gives Interact

Anthropic restricts what its most capable models can do in cybersecurity. Its Cyber Verification Program reduces those safeguards for verified security organizations, in three tiers: Defense Access, Red Team Access and Specialized Access. Each tier requires proof of security controls before access is granted.

As a Defence Access CVP participant, Interact’s security team now has access to Claude Mythos 5.1, Anthropic’s most powerful model for cyber work, and we’re pointing it directly at ourselves.

Claude Mythos 5.1 gives Interact’s security team an attacker that thinks like the most sophisticated adversaries our customers will ever face, and we are using it to put the Interact platform under relentless pressure so weaknesses are found and fixed by us, not by them.

“The intranet is now where AI meets your whole organization, and attackers have AI too. We’re not going to wait to find out what a frontier model can do to our platform. We’re finding out first, on our terms, and closing every door it opens.”

Tom Walters VP product, Interact Software

Why the intranet is a target

More than 1,000 organizations run their intranet on Interact, and for each of them it holds the map of the organization: its people, its structure, its policies and its knowledge. It’s connected to identity, HR and Microsoft 365. And it’s the most trusted and reliable voice the organization has.

In the wrong hands, that’s not just data. It’s a target list and a megaphone. An attacker who controls the intranet doesn’t need to fake a message from the CEO. They can send one. Now add AI. With Cue, Conductor and Interact MCP, assistants including Microsoft 365 Copilot and Claude can work directly with intranet content and actions. Every connection is a new door. Interact’s rule for AI on the intranet is simple: an agent must never be able to do more than the person it works for. Claude Mythos 5.1 lets us test that rule against the most capable attacker available, instead of hoping it holds.

What this means for Interact customers

Interact customers get a platform tested against the kind of attacker most security teams can only imagine.

• Tested at AI speed. Mythos can explore in hours what would take a human team weeks.
• Cyber defence of environments. The most powerful models supporting development of the strongest protection.
• Fixed before it matters. Anything we find goes straight into our security release process.
• Kept quiet where it counts. We won’t publish specific findings that could help anyone else.
• Your data stays yours. All testing runs in Interact-controlled environments, never on customer tenants or data.

Security teams who want to know more can talk to their Customer Success Manager or visit our Trust Center.

Looking ahead

Enterprise AI is only as trustworthy as the foundations it runs on. With Interact MCP, Cue, and Conductor, we intend to be the most secure foundation an AI-powered organization can build on.

“A secure intranet in 2026 means something completely different to what it meant even two years ago,” said Tom Walters, VP of Product at Interact Software. “The intranet is now where AI meets your whole organization, and attackers have AI too. We’re not going to wait to find out what a frontier model can do to our platform. We’re finding out first, on our terms, and closing every door it opens.”