Artificial intelligence agents are no longer a frontier experiment. They are the new workforce—and the new threat actor.
In April, Anthropic disclosed Claude Mythos Preview, a model so capable at offensive cyber work that the company refused to release it publicly. Anthropic engineers with no formal security training asked the model to find remote code execution vulnerabilities overnight. They woke up to complete, working exploits.
The results explain Anthropic's restraint. Mythos identified thousands of previously unknown zero-day vulnerabilities across every major operating system and every major web browser. It produced working exploits on the first attempt in more than 83% of cases. It found a flaw in OpenBSD (an operating system famous for its security hardening) that had survived 27 years of human review.
Anthropic's own team estimates similar capabilities will proliferate to other labs within six to eighteen months. Open-weight models will follow. That is the window defenders have—and it is narrower than most boardrooms yet appreciate.
Unfortunately, the security industry has been organized around a simple premise for two decades: the best-defended systems are safe because finding real vulnerabilities in them is hard, slow, and expensive. Mythos exposed that premise. The vulnerabilities were always there.
What changed is the cost of finding them, which is now effectively zero.
As our CEO Bipul Sinha recently stated, "Cybersecurity as we know it today is dead." Legacy systems were built for human-powered attacks. They cannot scale to machine speed. Fortunately, an intelligent recovery infrastructure can provide a last line of defense against AI-powered cyber escalation.
The Death of Detection
For twenty years, the security industry organized itself around a quiet assumption: detection buys you time.
Mythos ended that assumption. When an AI agent can read a codebase, hypothesize a flaw, build a proof of concept, and chain exploits together faster than an analyst can triage a ticket, detection only tells you what has already happened. The dwell time that made detection economically valuable has collapsed toward zero.
One industry study found that more than 88% of discovered vulnerabilities in large organizations still sit unpatched more than six months after disclosure. That was already a problem when the threat landscape moved at human speed. In a world where attackers can weaponize a newly published CVE within hours of the patch dropping, unpatched-at-twelve-months is not a backlog. It is a breach waiting for a timestamp.
Defense must shift, permanently, to an assume breach default. That means:
Assume a privileged account with weak MFA has been commandeered.
Assume a known vulnerability, if still unpatched after 72 hours, has been exploited.
Assume that a misconfigured, internet-connected server has already been compromised.
If you want to seriously embrace this posture, you must invest in three key strategies that will keep your enterprise safe, even after a successful incursion—remediation, recovery, and resilience.
Prevention Is a Crowded Graveyard
Indeed, recovery is the last line of defense and increasingly the only line that holds when attackers move at machine speed. But traditional recovery is a bottleneck. It starts with uncertainty—spending days rehydrating data just to find a clean restore point and weeks mapping out manual sequences.
When the window between disclosure and exploitation shrinks from months to minutes, the value of knowing exactly where your clean state lives—and being able to return to it fast—becomes the defining capability of a modern security program.
Rubrik was built for this moment, even before Mythos made it undeniable. The Rubrik Platform, powered by our Preemptive Recovery Engine™, is the industry's only natively-built solution that sits at the intersection of data, identity, and AI, designed to begin recovery before the attack even starts.
Rubrik’s Preemptive Recovery Engine™ leverages a continuous metadata layer, moving intensive analysis out of the critical path and performing the investigation before the incident even hits.
This architecture is designed for speed, turning weeks of downtime into minutes of recovery with:
Pre-computed clean states: The solution continuously calculates safe recovery points so teams can restore in up to 60 seconds—not the 50+ days of traditional scanning.
Unified metadata intelligence: Versioning, anomaly detection, and identity attributes are continuously logged, enabling cross-domain analysis in minutes, not weeks.
Automated recovery orchestration: Pre-defined plans automate boot order, IP configuration, and dependencies to bring applications back online systematically.
Agentic governance: Real-time monitoring of AI agents with policy guardrails—cutting them off when they cross boundaries.
The Result? Our Preemptive Recovery Engine™ collapses recovery timelines from weeks to hours. That way, you don’t just recover—you outrun the attack.
Surviving in a Post-Mythos World
Surviving in a post-Mythos world demands three architectural commitments:
Data Protection: You must be able to secure and recover the complete data estate, including the data inside AI models, embeddings, and vector stores. These are now high-value targets in their own right. A poisoned vector store is a quiet, persistent compromise that no EDR product was designed to see.
Identity Security: Every modern breach is an identity breach. The speed at which an enterprise can recover a clean, functional identity fabric is the difference between a bad week and a company-defining crisis. Ransomware groups already understood this. AI-assisted attackers will exploit it faster.
Visibility into Agentic Actions: Autonomous agents now execute critical business tasks inside enterprises. Governing those actions, auditing them, and rolling them back under fire is the new core of enterprise security. An agent with legitimate credentials doing illegitimate work is the attack pattern the next decade will be written about.
Rubrik Forward 2026
The threat model changed faster than most security strategies anticipated. Rubrik Forward 2026 is where the response model gets built.
Join us June 8-11 at The Venetian Resort in Las Vegas for sessions built for the leaders who have to make the hard calls. Talk with other CEOs, CIOs, and CISOs who need a resilience architecture that holds up—not in a vendor demo, but at 2 a.m. when the breach is live and the board is on the phone.
Register at rubrik.com/forward.