Picus Is Now Anthropic-Verified, With Mythos-Class AI Behind Its Autonomous Pentesting

Picus is verified under Anthropic's Cyber Verification Program, the pathway Anthropic is using to put Mythos-class cyber capabilities in defenders' hands. That power now runs inside Picus Autonomous Pentesting.

The techniques a defender uses to prove an exploit is real are the same ones an attacker uses to build it. Recon, exploitation, lateral movement, and others. None of it changes based on who's at the keyboard. That's the hard truth about pointing frontier AI at offensive security, and it's exactly why a general-purpose model refuses this work out of the box.

Picus is verified under Anthropic's Cyber Verification Program, the pathway Anthropic is using to extend Mythos-class cyber capabilities to defenders. Access to the Anthropic's most advanced frontier model for cybersecurity is tightly restricted because of how powerful it is on cyber tasks. Picus Autonomous Pentesting now runs with that access.

What Mythos-Class Access Really Gets You

When Anthropic laid out the expansion of Project Glasswing, it said plainly where the Cyber Verification Program is headed: scale it up to grant Mythos-class capabilities to many more organizations for specific cyberdefense tasks, with penetration testing named as one of them. Project Glasswing is Anthropic's effort to secure the world's most important software, with Mythos Preview at its core. And I've written about the paradox underneath all of it: the same model that lets defenders audit critical code hands attackers a faster way to weaponize it.

Resolving that paradox in the customer's favor is the entire point of Picus Autonomous Pentesting. This isn't a "we use AI now" announcement. It's verified access to frontier-grade exploitation, aimed at the one job the product exists to do.

The Exposure Decision Gap Nobody Else Closes

Finding exposure was never the hard part. Scanners have buried teams in vulnerability lists for twenty years. The hard part is proving the right call: patch, mitigate, monitor, or accept.

That's the exposure decision gap, and it has three roots. Exposure accumulates faster than anyone can safely clear it, so the backlog grows and the window stays open. Teams know which vulnerabilities exist, but not whether the exploit chain would actually succeed in their environment, against their controls. And when the decision finally gets made, it rests on severity scores, assumptions, and attestations. Nobody can prove it was defensible.

Picus Autonomous Pentesting answers the only question that ends that argument: can an attacker actually reach something that hurts by exploiting exposures, in your environment, right now.

How the AI Agents Actually Break In

Picus Autonomous Pentesting runs on AI agents that find, exploit, and prove what matters, then loop until the environment is clean. They plan the engagement against your crown jewels (the assets you can least afford to lose, like a customer database, domain admin, or a payments system) and your scope, map the full attack surface across external assets, internal network, and identities, then chain and execute real exploitation across hosts to capture concrete proof of compromise. From there they rank everything by blast radius (how far an attacker could spread from a single foothold) and fire the response, opening a ticket, triggering an automated response playbook, or pushing an indicator, signature, or detection rule straight to your defenses. Recon, exploit, prove, prioritize, on repeat.

Frontier access changes the part that always mattered most. The result is three things competitors keep promising and rarely deliver. It proves which exposures attackers can actually reach and weaponize, not another scanner finding list. It runs at machine speed, compressing weeks of pentesting into minutes, continuously and on demand. And it reveals the critical attack paths, mapping the full chain from foothold to crown jewels, because chained vulnerabilities cause breaches and isolated flaws mostly don't.

Beyond Autonomous Pentesting: The Full Validation Cycle

Picus Autonomous Pentesting doesn't fight alone. It's one part of a continuous validation cycle Picus delivers end to end. It answers whether exploitation succeeds in your environment. Picus Exposure Validation turns that finding into a defensible decision by testing whether an attacker's full sequence of techniques, the TTP chain (the tactics, techniques, and procedures they string together to reach a goal), beats your controls. Picus Breach and Attack Simulation (Security Control Validation) then proves whether those controls actually detect, block, and respond, and re-tests after every decision so the finding never goes stale. When a vulnerability lands with no safe public exploit, the cycle still works: map that vulnerability to its TTP chain, validate the behaviors against your real controls, and exploitability is proven the day it drops rather than guessed from a severity score.

Staying Ahead of What's Coming

Anthropic has been blunt about what's coming. Within a year, cheap, fast models with serious cyber capability will be everywhere, and plenty will ship with no guardrails at all. The breach you read about next year is being built right now, by an attacker with better tooling than your security team has ever had.

Unless your defender has better tooling still. That is the entire point of verified frontier access: a Mythos-class attacker working on your side, exploiting the attack paths before a real one finds them, proving every exposure that matters before it costs you anything.

The next attack will be AI-driven. Picus already is.