CrowdStrike deploys GPT-5.6 Cyber in a defensive harness and lands on the Claude Marketplace

CrowdStrike is betting that the same architecture that makes AI attackers faster can make defenders faster too. The company announced at Fal.Con that it will run OpenAI's GPT-5.6 Cyber inside a purpose-built cyber harness for risk assessment, while also placing its Falcon platform on Anthropic's Claude Marketplace with a novel procurement mechanic. For builders, the story is about the system, not just the model.

CrowdStrike runs GPT-5.6 Cyber inside a defensive harness

CrowdStrike's Frontier AI Readiness and Resilience service applies GPT-5.6 Cyber within CrowdStrike's purpose-built cyber harness. The stated jobs: assessing risk, analyzing attack paths, and setting remediation priorities under human oversight for approved defensive use only. The word "harness" is doing heavy lifting here.

A day earlier, Booz Allen's Cyber Weapon Index tested 18 models as autonomous attackers and found that the harness can matter as much as the model itself, or more. Claude Sonnet 5 finished 15th of 18 with a score of 13; fitted with a harness, it rivaled the leader at 80. The index's conclusion: the model is no longer the unit of risk, the system is. CrowdStrike is now selling that same architecture pointed defensively. But nobody has published a test of the defensive half.

Policing Codex agents at runtime

The other half of the OpenAI partnership is Falcon Guardian, CrowdStrike's AI detection and response product, applied to OpenAI's Codex agents. Falcon Guardian promises a live inventory of every Codex agent running in an organization, showing who deployed each one and what it can reach. It then watches those agents in real time and flags unauthorized behavior. Administrators define which actions are permitted. The pitch assumes enterprises are already running coding agents they cannot fully see, and that inventory is the first feature needed.

The Claude Marketplace procurement mechanic

Hours later, CrowdStrike announced a second deal with Anthropic. The Falcon platform is going onto the Claude Marketplace, and Anthropic customers will be able to buy it using part of the spending they have already committed to Anthropic. This is a procurement mechanic rather than a product: committed AI spend becomes a currency for buying unrelated enterprise software. Daniel Bernard, CrowdStrike's chief business officer, said AI is changing how technology is procured as well as how it is run.

The technical part is Charlotte AI AgentWorks. Security teams describe an outcome in plain language, the system builds an agent grounded in Falcon data, and the agent runs from inside Claude. Ash Alhashim, who leads enterprise cybersecurity sales at Anthropic, said customers get a platform they trust on commitments they have already made.

Agent identity and endpoint controls

CrowdStrike also introduced an Agentic Identity Provider, which issues AI agents their own identities and treats that as the control plane for the enterprise. A separate product blocks malicious open-source packages at the endpoint before their code runs. Together, these announcements describe a company that now sells the agents, the identity layer those agents authenticate through, the system that investigates them, and the frontier model that reasons about the risk they create.

What's missing and what to watch

OpenAI's Astra, which the company said had reached its critical cybersecurity capability threshold, was not part of either announcement and was not in the Booz Allen index. The index also found that a cyber-tuned model complied with a task that its non-tuned sibling refused, suggesting guardrails belong to the configuration rather than to the model. CrowdStrike's release does not address how it prevents that compliance gap in its own harness. Booz Allen expects most models to match the leader within six months. CrowdStrike is betting that defensive harnesses can outpace attacker-enabled models over time, but no independent test of the defensive half exists yet.